Patent Application
20120327779
The present invention generally relates to the field of communication systems and to systems and methods for congestion detection and packet characteristics detection for prioritizing and scheduling packets in a communication network.
In a communication network, such as an Internet Protocol (IP) network, each node and subnet has limitations on the amount of data which can be effectively transported at any given time. In a wired network, this is often a function of equipment capability. For example, a Gigabit Ethernet link can transport no more than 1 billion bits of traffic per second. In a wireless network the capacity is limited by the channel bandwidth, the transmission technology, and the communication protocols used. A wireless network is further constrained by the amount of spectrum allocated to a service area and the quality of the signal between the sending and receiving systems. Because these aspects can be dynamic, the capacity of a wireless system may vary over time.
Additionally, each node has limitations on the processing in can perform. Increasing the processing available may be expensive or may require the node to be taken out of service. Furthermore, a node may have many different functions that compete for the available processing. Even when sufficient processing ability is available, its use carries a cost, for example, in power consumption.
Systems and methods for providing parameterized (or weight-based) scheduling systems, with congestion detection are provided. In an embodiment, a method for operating a communication device for scheduling transmission of data packets is provided. The method includes: receiving data packets from a communication network; monitoring one or more connections associated with the received data packets to detect characteristics of the connections; inserting each of the data packets into one of a plurality of data queues; detecting information about congestion effecting communication of the data packets; determining scheduler parameters for the data queues, the scheduler parameters including factors based on the detected information about congestion and the detected characteristics associated with the data packets in the corresponding data queues; scheduling the data packets from the data queues for transmission taking into account the scheduler parameters; and transmitting the data packets based on the scheduling.
In an embodiment, a method for operating a communication device for scheduling transmission of data packets is provided. The method includes: receiving data packets from a communication network; monitoring one or more connections associated with the received data packets to detect characteristics of the connections; inserting each of the data packets into one of a plurality of data queues; calculating one or more metrics indicative of quality of experience (QoE) using the detected characteristics of the connections; determining scheduler parameters for the data queues, the scheduler parameters including factors based on the calculated metrics and the detected characteristics associated with the data packets in the corresponding data queues; scheduling the data packets from the data queues for transmission taking into account the scheduler parameters; and transmitting the data packets based on the scheduling.
In an embodiment, a communication device is provided. The communication device includes: a receiver module configured to receive data packets from a communication network; a packet inspection module configured to analyze the received data packets to determine which of the received data packets should be further inspected, detect information about connections used in transporting the data packets, detect information about streams, sessions, and applications associated with the data packets; and a processor module configured to detect information about congestion effecting communication of the data packets.
In an embodiment, a communication device is provided. The communication device includes: a receiver module configured to receive data packets from a communication network; a packet inspection module configured to analyze the received data packets to determine which of the received data packets should be further inspected, detect information about connections used in transporting the data packets, detect information about streams, sessions, and applications associated with the data packets; and a processor module configured to calculate one or more metrics indicative of quality of experience (QoE) based on the detected characteristics of the connections.
Other features and advantages of the present invention should be apparent from the following description which illustrates, by way of example, aspects of the invention.
The details of the present invention, both as to its structure and operation, may be gleaned in part by study of the accompanying drawings, in which like reference numerals refer to like parts, and in which:
Systems and methods for providing a parameterized scheduling system that incorporates end-user application awareness are provided. The systems and methods disclosed herein can be used with scheduling groups that contain data streams from heterogeneous applications. Some embodiments use packet inspection to classify data traffic by end-user application. Individual data queues within a scheduling group can be created based on application class, specific application, individual data streams or some combination thereof. Embodiments use application information in conjunction with Application Factors (AF) to modify scheduler parameters, thereby differentiating the treatment of data streams assigned to a scheduling group. In an embodiment, a method for adjusting the relative importance of different user applications through the use of dynamic AF settings is provided to maximize user QoE in response to recurring network patterns, one-time events, or both. In an embodiment, a method for maximizing user QoE for video applications by dynamically managing scheduling parameters is provided. This method incorporates the notions of “duration neglect” and “recency effect” in an end-user's perception of video quality (i.e. video QoE) in order to optimally manage video traffic during periods of congestion.
The systems and methods disclosed herein can be applied to various capacity-limited communication systems, including but not limited to wireline and wireless technologies. For example, the systems and methods disclosed herein can be used with Cellular 2G, 3G, 4G (including Long Term Evolution (“LTE”), LTE Advanced, WiMax), WiFi, Ultra Mobile Broadband (“UMB”), cable modem, and other wireline or wireless technologies. Although the phrases and terms used herein to describe specific embodiments can be applied to a particular technology or standard, the systems and methods described herein are not limited to these specific standards.
In office building 120(2), enterprise femtocell 140 provides in-building coverage to subscriber stations 150(3) and 150(6). Enterprise femtocell 140 can connect to core network 102 via ISP network 101 by utilizing broadband connection 160 provided by enterprise gateway 103.
Data networks (e.g. IP), in both wireline and wireless forms, have minimal capability to reserve capacity for a particular connection or user, and therefore demand may exceed capacity. This congestion effect may occur on both wired and wireless networks.
During periods of congestion, network devices must decide which data packets are allowed to travel on a network, i.e., which traffic is forwarded, delayed, or discarded. In a simple case, data packets are added to a fixed length queue and sent on to the network as capacity allows. During times of network congestion, the fixed length queue may fill to capacity. Data packets that arrive when the queue is full are typically discarded until the queue is drained of enough data to allow queuing of more data packets. This first-in-first-out (FIFO) method has the disadvantage of treating all packets with equal fairness, regardless of user, application, or urgency. This is an undesirable response as it ignores that each data stream can have unique packet delivery requirements, based upon the applications generating the traffic (e.g. voice, video, email, internet browsing, etc.). Different applications degrade in different manners and with differing severity due to packet delay and/or discard. Thus, a FIFO method is said to be incapable of managing traffic in order to maximize an end user's experience, often termed Quality of Experience (QoE).
In response, technologies have been developed to categorize packets and to treat data streams with differing levels of importance and/or to manage to differentiated levels of service. A data stream may be a stream of related packets from a single user application, for example, video packets of a YouTube video or the video packet portion of a video Skype session.
The processor module 281 is configured to process communications being received and transmitted by the station 277. The storage module 283 is configured to store data for use by the processor module 281. In some embodiments, the storage module 283 is also configured to store computer readable instructions for accomplishing the functionality described herein with respect to the station 277. In one embodiment, the storage module 283 includes a non-transitory machine readable medium. For the purpose of explanation, the station 277 or embodiments of it such as the base station, subscriber station, and femto cell, are described as having certain functionality. It will be appreciated that in some embodiments, this functionality is accomplished by the processor module 281 in conjunction with the storage module 283 and transmitter receiver module 279.
Many different nodes in a network (e.g., application server, proxy server, transport device such as a network switch or router, storage device, end-user device such as a smart phone, tablet, or laptop) may initiate or participate in a session. Nodes may host one or more sessions simultaneously. The simultaneous sessions may be independent from one another (e.g., a user using Facebook and email simultaneously) or related to each other (e.g., a browsing session which spawns two video streaming sessions). A session may be established between two nodes. Alternatively, sessions may be viewed as a relationship between one node and many nodes, for example, through the use of multicast and broadcast protocols.
Sessions may be characterized or categorized by various criteria. One criterion is the specific application (for example, the application program or software 1410) that was initiated by the user and was responsible for launching the session. Examples of specific applications include a YouTube app, a Chrome internet browser, and a Skype voice calling program. Another criterion is the application class that describes the overall function served by a particular session. Example application classes include streaming video, voice calling, internet browsing, email, and gaming.
A stream layer 1430 is the layer at which individual data streams that make up the session exist. A session may consist of one or more independent data streams using the same or potentially different underlying connections. For example, a single VoIP phone call session may contain two data streams. One data stream may serve the bidirectional voice traffic (which may be payload or data plane packets) using a User Datagram Protocol (UDP) connection. A second data stream may use one or more Transmission Control Protocol (TCP) connections to handle call setup/teardown (which may be signaling or control plane packets), as for example when using the session initiation protocol (SIP). In another example, for a video Skype call, there may be one stream to carry SIP signaling, to start, stop, and otherwise control the session, a second stream carrying voice packets using the Real-Time Transport (RTP) protocol, and a third stream carrying video packets using the RTP protocol.
A connection layer 1440 is the layer where the stream layer 1430 data is transported over some logical link provided by a logical link layer 1450. The connection layer 1440 protocols are neither application specific nor physical medium specific. A connection may refer to the underlying protocols used to transport session data and messages and to the group of packets, messages, and transactions used to establish (initiate) or remove (terminate) the connection. For example, a connection-oriented socket may be established via TCP between two nodes of an Internet Protocol (IP) network using a combination of IP addresses and port numbers. Once established, this TCP connection may be used to transport packets, for example, packets of a hyper-text transport protocol (HTTP) streaming video session. In an alternative to a TCP connection, a datagram socket can be established to transport traffic using UDP.
In the video Skype example, at the connection layer 1440, a SIP signaling stream 1432 is transported over a TCP/IP connection identified by source and destination IP addresses and TCP ports while a voice stream 1434 and a video stream 1436 are each transported over UDP/IP connections identified by source and destination IP addresses and UDP ports. While the UDP protocol is considered connectionless, it is convenient to use the term connection to also describe the UDP mechanisms that ensure the transport of data packets from the data source to the data sink for a stream.
The logical link layer 1450 is the layer at which a logical link exists that abstracts the actual physical medium and its transport mechanisms from the layers above. For example, in an LTE system, multiple connections (each carrying a stream) of the video Skype session are carried within an LTE data radio bearer (DRB) (for example, over wireless link 190 of
One method to assign importance and to optimize resource allocation between different data streams is through the use of desired performance requirements. For example, performance requirements may include desired packet throughput, and tolerated latency and jitter. Such performance requirements may be assigned based upon the type of data or supported application. For example, a voice over internet protocol (VoIP) phone call may be assigned the following performance requirements suited for the packet based transmission of voice through an IP network: throughput=32 kilobits per second (kbps), maximum latency=100 milliseconds (ms), and maximum jitter=10 ms. In contrast, a data stream which carries video may require substantially more throughput, but may allow for slightly relaxed latency and jitter performance as follows: throughput=2 megabits per second (Mbps), maximum latency=300 ms, maximum jitter=60 ms.
Scheduling algorithms located at network nodes can use these performance requirements to make packet forwarding decisions in an attempt to best meet each stream's requirements. The sum total of a stream's performance requirements is often described as the quality of service, or QoS, requirements for the stream.
Another method to assign importance is through the use of relative priority between different data streams. For example, standards such as the IEEE 802.1p and IETF RFC 2474 Diffsery define bits within the IP frame headers to carry such priority information. This information can be used by a network node's scheduling algorithm to make forwarding decisions, as is the case with the IEEE 802.11e wireless standard. Additional characteristics of a packet or data stream can also be mapped to a priority value, and passed to the scheduling algorithm. The standard 802.16e, for example, allows characteristics such as IP source/destination address or TCP/UDP port number to be mapped to a relative stream priority while also considering performance requirements such as throughput, latency, and jitter.
In some systems, data streams may be assigned to a discrete number of scheduling groups, defined by one or more common characteristics of scheduling method, member data streams, scheduling requirements or some combination thereof.
For example, scheduling groups can be defined by the scheduling algorithm to be used on member data streams (e.g., scheduling group #1 may use a proportional fair algorithm, while scheduling group #2 uses a weighted round-robin algorithm).
Alternatively, a scheduling group may be used to group data streams of similar applications (e.g., voice, video or background data). For example, Cisco defines six groups to differentiate voice, video, signaling, background, and other data streams. This differentiation of application may be combined with unique scheduling algorithms applied to each scheduling group.
In another example, the Third Generation Partnership Program (3GPP) has established a construct termed QoS Class Identifiers (QCI) for use in the Long Term Evolution (LTE) standard. The QCI system has 9 scheduling groups defined by a combination of performance requirements, scheduler priority and user application. For example, the scheduling group referenced by QCI index=1 is defined by the following characteristics:
The term ‘class of service’ (or CoS) is sometimes used as a synonym for scheduling groups.
In systems as described above, one or more data streams can be assigned an importance and a desired level of performance. This information may be used to assign packets from each data stream to a scheduling group and data queue. A scheduling algorithm can also use this information to decide which queues (and therefore which data streams and packets) to treat preferentially to others in both wired and wireless systems.
In some scheduling algorithms the importance and desired level of service of each queue is conveyed to the scheduler through the use of a scheduling weight. For example, weighted round robin (WRR) and weighted fair queuing (WFQ) scheduling methods both use weights to adjust service among data queues. In some scheduling algorithms the importance and desired level of service of each queue is conveyed to the scheduler through the use of credits and debits. For example, a proportional fair scheduler (PFS) method may use credits and debits to adjust service among data queues. Some algorithms use weights and convert them to credits in the form of number of packets or bytes to be served during a scheduling round.
In WRR, all non-empty queues are serviced in each scheduling round, with the number of data packets served from each queue being proportional to the weight of the queue. The weights may be derived from a variety of inputs such as relative level of service purchased (e.g., gold, silver, or bronze service), minimum guaranteed bit rates (GBR), or maximum allowable bit rates. In one example, three queues may have data pending. The queue weights are 1, 3, and 6 for queues 1, 2, and 3 respectively. If 20 packets are to be served during each round, then queues 1, 2, and 3 would be granted 10%, 30%, and 60% of the 20 packet budget or credits of 2, 6, and 12 packets, respectively. One skilled in the art will recognize that other weights can be applied as well and the concepts of weights, credits, and rates can be interchanged.
The WFQ algorithm is similar to WRR in that weighted data queues are established and serviced in an effort to provide a level of fairness across data streams. In contrast to WRR, WFQ serves queues by looking at number of bytes served, rather than number of packets. WFQ works well in systems where data packets may be fragmented into a number of pieces or segments, such as in WiMAX systems. In the example where three queues have data pending with queue weights 1, 3 and 6 for queues 1, 2 and 3 respectively, the weights would translate to credits of 10%, 30%, and 60% of the bandwidth available during that scheduling round.
The PFS algorithm typically uses a function of rates such as GBR or maximum allowable rates to directly calculate credits each queue receives each scheduling round. For example, if a service is allowed a rate of 768 kilobytes per second, and there are 100 scheduling rounds per second, the service's queue would receive a credit of 7680 bytes per scheduler round. The amount actually allocated to the queue during a scheduler round is debited from the queue's accumulated credit. Credits can be adjusted or accumulated, round-by-round, in an effort to balance the performance requirements of multiple queues. For example, a first queue which has been allocated resources below its minimum GBR specification may have accumulated credits (typically up to some allowable cap) effectively causing its weight to increase in relation to a second queue which has been allocated capacity substantially above its GBR, effectively causing the second queue to accumulate a negative credit, or debit.
Input traffic 305 can consist of a heterogeneous set of individual data streams each with unique users, sessions, logical connections, performance requirements, priorities, or policies that enter the scheduling system. Classification and queuing module 310 is configured to assess the relative importance and assigned performance requirements of each packet and to assign the packet to a scheduling group and data queue. According to an embodiment, the classification and queuing module 310 is configured to assess the relative importance and assigned performance requirements of each packet using one of the methods described above, such as 802.1p or Diffserv.
According to an embodiment, the parameterized scheduling system 300 is implemented to use one or more scheduling groups and each scheduling group may have one or more data queues associated with the group. According to an embodiment, each scheduling group can include a different number of queues, and each scheduling group can use different methods for grouping packets into queues, or a combination thereof. A detailed description of the mapping between input traffic, scheduling groups, and data queues is presented below.
According to an embodiment, classification and queuing module 310 outputs one or more data queues 315 and classification information 330 which is received as an input at scheduler parameter calculation module 335. The phrase “outputs one or more data queues” is intended to encompass populating the data queues and does not require actual transmission or transfer of the queues. According to an embodiment, the classification information 330 can include classifier results, packet size, packet quantity, and/or current queue utilization information. Scheduler parameter calculation module 335 is configured to calculate new scheduler parameters (e.g., weights and/or per scheduler round credits) on a per queue basis. Scheduler parameter calculation module 335 can be configured to calculate the new parameters based on a various inputs, including the classification information 330, optional operator policy and service level agreement (SLA) information 350, and optional scheduler feedback information 345 (e.g., stream history received or resource utilization from scheduler module 320). Scheduler parameter calculation module 335 can then output scheduler parameters 340 to one or more scheduler modules 320.
Scheduler module 320 receives the scheduler parameters 340 and the data queues 315 (or accesses the data queues) output by classification and queuing module 310. Data queues as described herein can be implemented in various ways. For example, they can contain the actual data (e.g., packets) or merely pointers or identifiers of the data (packets). Scheduler module 320 uses the updated scheduler parameters 340 to determine the order in which to forward packets (or fragments of packets) from the data queues 315 to output queue 325, for example using one of the methods described above such as PFS, WRR or WFQ. In an embodiment, the output queue 325 is implemented as pointers to the data queues 315. The traffic in the output queue 325 is de-queued and fed to the physical communication layer (or ‘PHY’) for transmission on a wireless or wireline medium.
Heterogeneous input traffic 305 is input into packet inspection module 410 which characterizes each packet to assess performance requirements and priority as described above. Based upon this information, each packet is assigned one of three scheduling groups 420, 425 and 430. While the embodiment illustrated in
In one example, an LTE eNB is configured to assign each QCI to a separate scheduling group (e.g., packets with QCI=9 may be assigned to one scheduling group and packets with QCI=8 assigned to a different scheduling group). Furthermore, packets with QCI=9 may be assigned to individual queues based on user ID, bearer ID, SLA or some combination thereof. For example, each LTE UE may have a default bearer and one or more dedicated bearers. Within the QCI=9 scheduling group, packets from default bearers may be assigned to one queue and packets from dedicated bearers may be assigned a different queue.
The method begins with receiving input traffic to be scheduled to be transmitted across a network medium (step 1205). According to an embodiment, the network medium can be a wired or wireless medium. According to an embodiment, the input traffic is input traffic 305 described above. The input traffic can consist of a heterogeneous set of individual data streams each associated with users, sessions, logical links, connections, performance requirements, priorities, or policies. According to an embodiment, classification and queuing module 310 can perform step 1205. According to an embodiment, packet inspection module 410 can perform this assessment step.
The input traffic can then be classified (step 1210). According to an embodiment, classification and queuing module 310 can perform step 1210. In this classification step, the input traffic is assessed to determine relative importance of each packet and to determine if performance requirements have been assigned for each data packet. For example, in an LTE network, a packet gateway can assign packets to specific logical link or bearers. This is indicated by assigning the same tunnel ID to packets for the same logical link (logical channel). The tunnel ID is mapped to an LTE scheduling group (i.e. QCI) when the logical bearer is established. This in turn implies certain performance requirements that are associated with the scheduling group. The tunnel ID may be detected and used to determine performance requirements and scheduling groups and to assign the packet to a queue. Similarly, in WiMAX, a service flow ID may be used for a similar purpose. According to an embodiment, packet inspection module 410 can perform this assessment step. This information can then be used by the classification and queuing module 310 to determine which scheduling groups the data packets should be added.
The input traffic can then be segregated into a plurality of scheduling groups (step 1215). The classification and queuing module 310 can use the information from the classification step to determine a scheduling group into which each data packet should be added. According to an embodiment, packet inspection module 410 of the classification and queuing module 310 can perform this step. According to an embodiment, the relative importance and assigned performance requirements of each packet is assessed using one of the methods described above, such as 802.1p or Diffserv.
The data packets comprising the input traffic can then be inserted into one or more data queues associated with the scheduling groups (step 1220). According to an embodiment, packet inspection module 410 of the classification and queuing module 310 can perform this step.
Scheduler parameters can then be calculated for each of the data queues (step 1225). According to an embodiment, this step is implemented by scheduler parameter calculation module 335. The scheduler parameters for each of the data queues is calculated based on the classification information created in step 1210. The classification information 330 can include classifier results, connection identifiers (e.g., source and destination IP address, TCP port, UDP socket), logical link identifiers (e.g., tunnel ID or bearer ID in LTE, service flow ID or connection ID in WiMAX), packet size, packet quantity, and/or current queue utilization information. The calculation of the scheduler parameters can also take into account other inputs including optional operator policy and service level agreement (SLA) information and optional scheduler feedback information.
Once the data packets have been added to the queues, data packets can be selected from each of the queues based on scheduler parameters (such as weights and credits) associated with those queues and inserted into an output queue (step 1230). The data packets in the output queue can then be de-queued and fed to the physical communication layer (or ‘PHY’) for transmission on a wireless or wireline medium (step 1235). According to an embodiment, scheduler module 320 can implement steps 1230 and 1235 of this method.
In WRR, WFQ, PFS or other weight or credit-based algorithms, some systems assign packets to queues and calculate scheduler parameters based on priority, performance requirements, scheduling groups, or some combination thereof. There are numerous deficiencies in these approaches.
For example, schedulers that consider performance requirements are typically complex to configure, requiring substantial network operator knowledge and skill, and may not be implemented sufficiently to distinguish data streams from differing applications. This leads to the undesirable grouping of both high and low importance data streams in a single queue or scheduling group. Consider, for example, an IEEE 802.16 network. Sometimes it is not possible or not practical to differentiate individual streams as described with reference to
There are numerous potential deficiencies of a priority-based weight or credit calculation system. The system used to assign priority may not be aware of the user application and in some cases cannot correctly distinguish among multiple data streams being transported to or from a specific user. The priority assignment is static and cannot be adjusted to account for changing network conditions. Priority information can be missing due to misconfiguration of network devices or even stripped due to network operator policy. The number of available priority levels can be limited, for example the IEEE 802.1p standard only allows 8 levels. In addition there can be mismatches due to translation discrepancies from one standard to another as packets are transported across a communication system.
A discrepancy between two different priority systems can exist in the example illustrated in
Some systems have combined the concepts of priority and performance requirements in an effort to provide additional information to the scheduling system. For example, in 802.16 the importance of streams (or “services”) is defined by a combination of priority value (based on packet markings such as 802.1p) and performance requirements. While a combined system such as 802.16 can provide the scheduler with a richer set of information, the deficiencies described above still apply.
The use of scheduling groups alone or in conjunction with the aforementioned techniques has numerous deficiencies in relation to end user QoE. For example, the available number of groups is limited in some systems which can prevent the fine-grained control necessary to deliver optimal QoE to each user. Additionally, some systems typically utilize a “best effort” group to describe those queues with the lowest importance. Data streams may fall into such a group because they are truly least important but also because such streams have not been correctly classified (intentionally or unintentionally), through the methods described above, as requiring higher importance.
An example of such a problem is the emergence of ‘over-the-top’ voice and video services or applications. These services provide capability using servers and services outside of the network operator's visibility and/or control. Data streams from an operator owned or sanctioned source, such as operator provided voice or video, may be differentiated onto different service flows, bearers (logical link), or connections prior to reaching a wireless access node such as a base station. This differentiation often maps to differentiation in scheduling groups and queues. However services, and the resultant data streams, from other sources may all be bundled together onto a default, often best effort, logical link or bearer. For example, Skype and Netflix are two internet-based services or applications which support voice and video, respectively. Data streams from these applications can be carried by the data service provided by wireless carriers such as Verizon or AT&T, to whom they may appear as non-prioritized data rather than being identified as voice or video. As such, the packets generated by these applications, when transported through the wireless network, may be treated on a ‘best-effort’ basis with no priority given to them above typical best-effort services such as web browsing, email or social network updates.
Some systems implement dynamic adjustment of scheduling weights or credits. For example, in order to meet performance requirements such as guaranteed bit rate (GBR) or maximum latency, scheduling weights may be adjusted upward or scheduling credits may accumulate for a particular data stream as its actual, scheduled throughput drops closer to the guaranteed minimum limit. However, this adjustment of weights or credits does not take into account the effect of QoE on the end user. In the previous example, the increase of weight or accumulation of credits to meet GBR limit may result in no appreciable improvement in QoE, yet create a large reduction in QoE for a competing queue with lower weight per scheduling round credit, or accumulated credit (or debit).
Therefore, there is a need for a system and method to improve the differentiation of treatment of data packets streams from heterogeneous applications grouped into the same scheduling group, such as is common for a ‘best effort’ scheduling group. Additionally, there is a need to extend the information provided to a parameterized scheduler beyond priority and performance requirements in order to maximize user QoE across a network.
As described above, communication systems can use classification and queuing methods to differentiate data streams based on performance requirements, priority and logical connections.
To address previously noted deficiencies in some systems, the classification and queuing module 310 of
Except as specifically noted, the elements of
Except as specifically noted, the elements of
According to an embodiment, the enhanced classification steps disclosed herein can be implemented in the enhanced packet inspection module 410′ of the enhanced classification and queuing module 310′. For example, 2-way video conferencing, unidirectional streaming video, online gaming, and voice are examples of some different application classes. Specific applications refer to the actual software used to generate the data stream traveling between source and destination. Some examples include: YouTube, Netflix, Skype, and iChat. Each application class can have numerous, specific applications. The table provided in
According to an embodiment, the enhanced classification and queuing module 310′ can inspect the IP source and destination addresses in order to determine the application class and specific application of the data stream. With the IP source and destination addresses, the enhanced classification and queuing module 310′ can perform a reverse domain name system (DNS) lookup or Internet WHOIS query to establish the domain name and/or registered assignees sourcing or receiving the Internet-based traffic. The domain name and/or registered assignee information can then be used to establish both application class and specific application for the data stream based upon a priori knowledge of the domain or assignee's purpose. The application class and specific application information, once derived, can be stored for reuse. For example, if more than one user device accesses Netflix, the enhanced classification and queuing module 310′ can be configured to cache the information so that the enhanced classification and queuing module 310′ would not need to determine the application class and specific application for subsequent accesses to Netflix by the same user device or another user device on the network.
For example, if traffic with a particular IP address yielded a reverse DNS lookup or WHOIS query which included the name ‘Youtube’ then this traffic stream could be considered a unidirectional video stream (application class) using the Youtube service (Specific Application). According to an embodiment, a comprehensive mapping between domain names or assignees and application class and specific application can be maintained. In an embodiment, this mapping is periodically updated to ensure that the mapping remains up to date.
According to another embodiment, the enhanced classification and queuing module 310′ is configured to inspect the headers, the payload fields, or both of data packets associated with various communications protocols and to map the values contained therein to a particular application class or specific application. For example, according to an embodiment, the enhanced classification and queuing module 310′ is configured to inspect the Host field contained in an HTTP header. The Host field typically contains domain or assignee information which, as described in the embodiment above, is used to map the stream to a particular application class or specific application. For example an HTTP header field of “v11.1scache4.c.youtube.com” could be inspected by the Classifier and mapped to Application Class=video stream, Specific Application=Youtube.
According to another embodiment, the enhanced classification and queuing module 310′ is configured to inspect the ‘Content Type’ field within a Hyper Text Transport Protocol (HTTP) packet. The content type field contains information regarding the type of payload, based upon the definitions specified in the Multipurpose Internet Mail Extensions (MIME) format as defined by the Internet Engineering Task Force (IETF). For example, the following MIME formats would indicate either a unicast or broadcast video packet stream: video/mp4, video/quicktime, video/x-ms-wm. In an embodiment, the enhanced classification and queuing module 310′ is configured to map an HTTP packet to the video stream application class if the enhanced classification and queuing module 310′ detects any of these MIME types within the HTTP packet.
In another embodiment, the enhanced classification and queuing module 310′ is configured to inspect a protocol sent in advance of the data stream. For example, the enhanced classification and queuing module 310′ may be configured to identify the application class or specific application based on the protocol used to set up or establish a data stream instead of identifying this information using the protocol used to transport the data stream. That is, the enhanced classification and queuing module 310′ may identify the application class or specific application by analyzing a stream of control packets rather than the information associated with connection layer 1440. According to an embodiment, the protocol sent in advance of the data stream is used to identify information on application class, specific application, and characteristics that allow the connection for transport of the data stream to be identified once initiated.
For example, in an embodiment, the enhanced classification and queuing module 310′ is configured to inspect Real Time Streaming Protocol (RTSP) packets which can be used to establish multimedia streaming sessions. RTSP packets are encapsulated within TCP/IP frames and carried across an IP network, as shown for an Ethernet based system in
RTSP (H. Schulzrinne, et al., IETF RFC 2326, Real Time Streaming Protocol (RTSP)) establishes and controls the multimedia streaming sessions with client and server exchanging the messages. An RTSP message sent from client to server is a request message. The first line of a request message is a request line. The request line is formed with the following 3 elements: (1) Method; (2) Request-URI; and (3) RTSP-Version.
RTSP defines methods including OPTIONS, DESCRIBE, ANNOUNCE, SETUP, PLAY, PAUSE, TEARDOWN, GET_PARAMETER, SET_PARAMETER, REDIRECT, and RECORD. Below is an example of a message exchange between a client (“C”) and a server (“S”) using method DESCRIBE. The response message from the server has a message body which is separated from the response message header with one empty line.
Request-URI in an RTSP message always contains the absolute URI as defined in RFC 2396 (T. Berners-Lee, et al., IETF RFC 2396, “Uniform Resource Identifiers (URI): Generic Syntax”). An absolute URI in an RTSP message contains both the network path and the path of the resource on the server. The following is the absolute URI in the message listed above.
rtsp://s.companydomain.com:554/dir/f.3gp
RTSP-Version indicates which version of the RTSP specification is used in an RTSP message.
In one embodiment, the enhanced classification and queuing module 310′ is configured to inspect the absolute URI in the RTSP request message and extract the network path. The network path typically contains domain or assignee information which, as described in the embodiment above, is used to map the stream to a particular application class or specific application. For example, an RTSP absolute URI “rtsp://v4.cache8.c.youtube.com/dir_path/video.3gp” could be inspected by the Classifier and mapped to Application Class=video stream, Specific Application=Youtube. In one embodiment, the enhanced classification and queuing module 310′ inspects packets sent from a client to a server to classify related packets sent from the server to the client. For example, information from an RTSP request message sent from the client may be used in classifying responses from the server.
The RTSP protocol may specify the range of playback time for a video session by using the Range parameter signaled using the PLAY function. The request may include a bounded (i.e.—start, stop) range of time or an open-end range of time (i.e. start time only). Time ranges may be indicated using either the normal play time (npt), smpte or clock parameters. Npt time parameters may be expressed in either hours:minutes:seconds.fraction format or in absolute units per ISO 8601 format timestamps. Smpte time values are expressed in hours:minutes:seconds.fraction format. Clock time values are expressed in absolute units per ISO 8601 formatted timestamps. Examples of Range parameter usage are as follows:
In one embodiment, the enhanced classification and queuing module 310′ is configured to inspect the RTSP messages and extract the Range information from a video stream using the npt, smpte, or clock fields. One skilled in the art would understand that the npt, smpte, and clock parameters within an RTSP packet may use alternate syntaxes in order to communicate the information described above.
The RTSP protocol includes a DESCRIBE function that is used to communicate the details of a multimedia session between Server and Client. This DESCRIBE request is based upon the Session Description Protocol (SDP is defined in RFC 2327 and RFC 4566 which supersedes RFC 2327) which specifies the content and format of the requested information. With SDP, the m-field defines the media type, network port, protocol, and format. For example, consider the following SDP media descriptions:
In the first example, an audio stream is described using the Real-Time Protocol (RTP) for data transport on Port 49170 and based on the format described in the RTP Audio Video Profile (AVP) number 0. In the second example, a video stream is described using RTP for data transport on Port 51372 based on RTP Audio Video Profile (AVP) number 31.
In both RTSP examples, the m-fields are sufficient to classify a data stream to a particular application class. Since the m-fields call out communication protocol (RTP) and IP port number, the ensuing data stream(s) can be identified and mapped to the classification information just derived. However, classification to a specific application is not possible with this information alone.
The SDP message returned from the server to the client may include additional fields that can be used to provide additional information on the application class or specific application.
An SDP message contains the payload type of video and audio stream transported in RTP. Some RTP video payload types are defined in RFC 3551 (H. Schulzrinne, et al., IETF RFC 3551, “RTP Profile for Audio and Video Conferences with Minimal Control”). For example, payload type of an MPEG-1 or MPEG-2 elementary video stream is 32, and payload type of an H.263 video stream is 34. However, payload type of some video codecs, such as H.264, is dynamically assigned, and an SDP message includes parameters of the video codec. In one embodiment, the video codec information may be used in classifying video data streams, and treating video streams differently based on video codec characteristics.
An SDP message may also contain attribute “a=framerate:<frame rate>”, which is defined in RFC 4566, that indicates the frame rate of the video. An SDP message may also include attribute “a=framesize:<payload type number><width><height>”, which is defined in 3GPP PSS (3GPP TS 26.234, “Transparent End-to-End Packet-switched Streaming Service, Protocols and Codecs”), may be included in SDP message to indicate the frame size of the video. For historical reasons, some applications may use non-standard attributes such as “a=x-framerate: <frame rate>” or “a=x-dimensions: <width><height>” to pass similar information as that in “a=framerate:<frame rate>” and “a=framesize:<payload type number><width><height>”. In one embodiment, the enhanced classification and queuing module 310′ is configured to inspect the SDP message and extract either the frame rate or the frame size or both of the video if the corresponding fields are present, and use the frame rate or the frame size or both in providing additional information in mapping the stream to a particular application class or specific applications.
In one embodiment, the enhanced classification and queuing module 310′ inspects network packets directly to detect whether these packets flowing between two endpoints contain video data carried using RTP protocol (H. Schulzrinne, et al., IETF RFC 3550, “RTP: A Transport Protocol for Real-Time Applications”), and the enhanced classification and queuing module 310′ performs this without inspecting the SDP message or any other message that contains the information describing the RTP stream. This may happen, for example, when either the SDP message or any other message containing similar information does not pass through the enhanced classification and queuing module 310′, or some implementation of the enhanced classification and queuing module 310′ chooses not to inspect such message. An RTP stream is a stream of packets flowing between two endpoints and carrying data using RTP protocol, while an endpoint is defined by a (IP address, port number) pair.
The RTP stream detection module 7110 parses the first several bytes of UDP or TCP payload according to the format of an RTP packet header and checks the values of the RTP header fields to determine whether the stream flowing between two endpoints is an RTP stream.
If a stream is detected to be an RTP stream, the video stream detection module 7120 will perform further inspection on the RTP packet header fields and the RTP payload to detect whether the RTP stream carries video and which video codec generates the video stream.
Payload type of some RTP payloads related to video is defined in RFC 3551. However, for a video codec with dynamically assigned payload type, the codec parameters are included in an SDP message. However, that SDP message may not be available to the video stream detection module 7120.
If the video stream detection module 7120 detects that payload type is dynamically assigned, it collects statistics regarding the stream. For example, statistics of values of the RTP header field “timestamp,” RTP packet size, and RTP packet data rate may be collected. The video stream detection module 7120 may then use one of the collected statistics or a combination of the statistics to determine whether the RTP stream carries video data.
A video stream usually has some well-defined frame rate, such as 24 FPS (frames per second), 25 FPS, 29.97 FPS, 30 FPS, or 60 FPS, etc. In one embodiment, the video stream detection module 7120 detects whether an RTP stream carries video data at least partially based on whether values of the RTP packet timestamp change in integral multiples of a common frame temporal distance (which is the inverse of a common frame rate).
A video stream usually has higher average data rate and larger fluctuation in the instantaneous data rate compared with an audio stream. In another embodiment, the video stream detection module 7120 detects whether an RTP stream carries video data at least partially based on the magnitude of the average RTP data rate and the fluctuation in the instantaneous RTP data rate.
The RTP payload format is media specific. For example, H.264 payload in an RTP packet always starts with a NAL unit header whose structure is defined in RFC 6814 (Y. K. Wang, et al., IETF RFC 6184, “RTP Payload Format for H.264 Video”). In one embodiment, the video stream detection module 7120 detects which video codec generates the video data carried in an RTP stream at least partially based on the pattern of the first several bytes the RTP payload.
According to an embodiment, the enhanced classification and queuing module 310′ can also be configured to implement enhanced queuing techniques. As described above, once enhanced classification has been completed, the enhanced classification and queuing module 310′ can assign to an enhanced set of queues based on the additional information derived by the enhanced classification techniques described above. For example, in an embodiment, the packets can be assigned to a set of queues by: application class, specific application, individual data stream, or some combination thereof.
In one embodiment, the enhanced classification and queuing module 310′ is configured to use a scheduling group that includes unique queues for each application class. For example, an LTE eNB may assign all QCI=6 packets to a single scheduling group. But with enhanced queuing, packets within QCI=6 which have been classified as Video Chat may be assigned to one queue, while packets classified as Voice may be assigned to a different queue, allowing differentiation in scheduling.
In another alternative embodiment, the enhanced classification and queuing module 310′ is configured to use a scheduling group that includes unique queues for each specific application. For example, an LTE eNB implementing enhanced queuing may assign QCI=9 packets classified as containing a Youtube streaming video to one scheduling queue, while assigning packets classified as a Netflix streaming video to a different scheduling queue. Even though they are the same application class, the packets are assigned different queues in this embodiment because they are different specific applications.
In yet another embodiment, the enhanced classification and queuing module 310 is configured such that a scheduling group may consist of unique queues for each data stream. For example an LTE eNB may assign all QCI=9 packets to a single scheduling group. Based on enhanced classification methods described above, each data stream is assigned a unique queue. For example, consider an example embodiment with a scheduling group servicing five mobile phone users, each running two specific applications. In one embodiment, if the applications for each mobile device are mapped to the default radio bearer for the mobile this would result in five queues, one for each mobile, carrying heterogeneous data using the original classification and queuing module. However, in one embodiment, ten queues are created by the enhanced classification and queuing module 310 in support of the ten data streams. In an alternative example, each of the five mobiles has two data streams which use the same specific application. In this case, the data streams are also classified based on, for example, port number or session ID into separate queues resulting in ten queues.
The enhanced categorization and queuing techniques described above can be used to improve the queuing in a wireless or wired network communication system. The techniques disclosed herein can be combined with other methods for assigning packets to queues to provide improved queuing.
According to an embodiment, the scheduler parameter calculation module 335 is configured to use enhanced policy information when calculating scheduler parameters to address QoE deficiencies of some weight or credit calculation techniques described above. According to an embodiment, the enhanced policy information 350 can include the assignment of a quantitative level of importance and relative priority based upon application class and specific application. This factor is referred to herein as the application factor (AF) and the purpose of the AF is to provide the operator with a means to adjust the relative importance, and ultimately the scheduling parameters, of queues following enhanced classification and enhanced queuing. In another embodiment, AFs are established through the use of internal algorithms or defaults, requiring no operator involvement.
Within the video chat class, the operator may discover that one video chat service (e.g., iChat) is substantially more burdensome (e.g., requires more capacity, has less latency or jitter tolerance) than another (e.g., Skype video), and can attempt to encourage the use of the more network friendly application by assigning a higher AF value to the Skype video chat than to iChat (8 versus 5).
Similarly, the operator may decide to preserve the QoE of a paid service, such as Netflix, at the expense of what may be considered the less important need to view short, free services, such as YouTube videos by adjusting the AF associated with these services. The operator may desire the ability to enhance certain voice services (e.g., Skype audio, Vonage) who have engaged strategically with the Operator with a high AF (8 and 6, respectively) while assigning all remaining (i.e. non-strategic) voice services a very low AF of 1.
One of ordinary skill in the art would understand that different AF values could be used to create different and varying weight or credit relationships between the application classes and specific applications. One skilled in the art would also understand how additional application classes and specific applications beyond those shown in
Additionally, one of ordinary skill in the art would understand that AFs may be assigned differently based upon node type and/or node location. For example, an LTE eNB serving a suburban, residential area may be configured to use one set of AFs while an LTE eNB serving a freeway may be configured use a different set of AFs.
According to an embodiment, enhanced scheduler parameter calculation module 335 can also be configured to implement enhanced techniques for determining weighting or credit factors. As described above, some weight or credit calculation algorithms can adjust scheduling parameters for individual queues based on various inputs. For example, in the parameterized scheduling module illustrated in
According to an embodiment, an enhanced scheduler parameter calculation module 335 can use additional weight and credit calculation factors to improve QoE performance. For example, in an embodiment, an additional weight factor can be used to generate an enhanced weight (W′) as shown below:
W′(q)=a*W(q)+b*AF(q)
where:
W′=enhanced queue weight
q=the queue index
W=the queue weight derived by conventional weight calculations
a=coefficient mapping W to W′
AF=the Application Factor
b=coefficient mapping AF to W′
For example, in an embodiment, an LTE eNB base station with 5 active streams (designated by a stream index i) within a single queue, best effort scheduling group (e.g., QCI=9 in LTE), is shown in
For example, stream #1, a Facebook request, and stream #4, a Skype video chat session, are both assigned to the same queue. Because packets from both streams are in the same queue, both streams must share the resources provided by the scheduler in a non-differentiated manner. For example, packets may be serviced in a FIFO method from the single queue thereby creating a “first to arrive” servicing of packets from both streams. This is undesirable during times of network congestion, due to the fact that a video chat session is more sensitive, in terms of user QoE, to packet delay or discard than a Facebook update.
In contrast, if the enhanced weight calculation technique described above (which can be implemented in enhanced scheduler parameter calculation module 335) are applied, each of the five streams (designated by index i in
Weights W1 and W2 are calculated for each stream using the equation for W′ (described above) with coefficient ‘a’ set to 1, and coefficient ‘b’ set to 0.5 and 1, respectively. That is:
W1(q)=W(q)+0.5*AF(q)
W2(q)=W(q)+AF(q)
The effect of the calculation can be seen by again comparing data stream #1 with stream #4. For W1, the video chat stream has a weight of 7 which is now larger than the Facebook stream weight of 4. As coefficient ‘b’ is increased to 1.0 in the calculations of W2, the difference in weight between stream #4 and #1 increases further (11 and 5, respectively).
For cases W1 and W2, the Skype stream will be allocated more resources than the Facebook stream. This increases the likelihood that the Skype session will be favored by the scheduler and can improve session performance and QoE during times of network congestion. While this comes at the expense of the Facebook session, the tradeoff is asymmetrical: packet delay/discard will have a smaller effect (i.e. less noticeable) on the Facebook session as compared to the equivalent packet treatment for a video chat session. Therefore the application-aware scheduling system has provided a more optimal response with respect to end-user QoE.
In an alternative example, each data stream in
Similarly, an enhanced per scheduling round credit could be calculated for credit-based scheduling algorithms using the formula C′(q)=a*C(q)+b*AF(q), where C (for credit) replaces the W (for weight) in the enhanced weight calculation formula. This enhanced credit would be added to the queue's accumulated credit (possibly capped) each scheduling round while allocated bandwidth would be debited from the accumulated credit. The AF is used in the same manner for both credit and weight based calculations, although the scale of AF may differ in the credit-based equation relative to the weight-based equation due to the typical difference in scale between weights and data rates when used in scheduling algorithms.
One of ordinary skill in the art would also recognize that the systems and methods described above may be extended to cases for which a queue contains packets from more than one data stream, more than one specific application, more than one application class, or combinations thereof for which an aggregate scheduling may be appropriate. For example, an enhanced weight or credit may be assigned to a queue containing three Skype/Video Chat data streams generated by three different mobile phones. Additionally, the systems and methods described above may be applied to all or only a subset of queues in one or more scheduling groups. For example, enhanced parameter calculation and enhanced queuing may be applied to an LTE QCI=9 scheduling group but known parameter calculation may be applied to LTE QCI=1-8 scheduling groups. Furthermore, the mapping of coefficients ‘a’ and ‘b’ may be adjusted as a function of scheduling group or alternative grouping of queues. For example, coefficient ‘b’ may be set to 1 for a scheduling group containing LTE QCI=9 queues but set to 0.5 for LTE QCI=8 queues.
According to an embodiment, the enhanced scheduler parameter calculation module 335 can also be configured to extend the application factor (AF) from a constant to one or more time-varying functions, AF(t). According to some embodiments, the AF is adjusted based upon a preset schedule. An operator may desire a particular treatment of applications at one time during the day and a differing treatment during other times.
For example, in one embodiment, the enhanced scheduler parameter calculation module 335 is configured to use “rush hour” AF values during typical commute times where voice calls are the predominant application running on a mobile network, especially for those cells and sectors serving transportation routes. For such times, (e.g., Monday through Friday, 7 am to 9 am and 4 pm to 7 pm) all voice applications are assigned an AF=10 improving the level of service above all other applications (referencing
In another example, the enhanced scheduler parameter calculation module 335 is configured to use larger AF values with over-the-top (OTT) video services during periods where such services are most likely to be used. For example, the enhanced scheduler parameter calculation module 335 is configured to use larger AF values during evenings on weekends, especially for networks that service residential areas. Referring once again to
The overall quantity of data for a particular application class or specific application can be used in the calculation and assignment of AFs. For example, if all data were from the same specific application, there may be no need to adjust AFs since all streams would warrant the equivalent user experience (however, even then characteristics, such as frames per second or data rate per stream, could still be used to modify AFs as described below). If there was very little data requiring a high quality of user experience, for example only one active Netflix session with all other data being email, the AF of the Netflix stream may be increased much more than would normally be the case to ensure the best quality of experience (for example, fewest lost packets) possible, knowing all or most other data is delay tolerant and may have built-in retransmission mechanisms. In an alternative embodiment, the AF is calculated as a function of the percentage of total available bandwidth required by homogenous or similar data streams. For example, Netflix streams could start with a high AF, but as a higher percentage of data usage is consumed by Netflix, the AF for all Netflix streams may decrease, or the AF for new Netflix streams may decrease leaving existing Netflix streams' AFs unchanged.
One of ordinary skill in the art would recognize that periodic, schedule based AF adjustments can be based on any recurring period including, but not limited to, time of day, day of week, tide, season and holidays. Furthermore, in an embodiment, the enhanced scheduler parameter calculation module 335 is configured to use non-recurring scheduling to adjust the AF in response to local sporting, business and community activities or other one-time scheduled events. According to some embodiments, the AF values can be manually configured by a network operator for non-recurring scheduling. According to other embodiments, the enhanced scheduler parameter calculation module 335 is configured to access event information stored on the network (or in some embodiments pushed to the network node on which the enhanced scheduler parameter calculation module 335 is implemented) and the enhanced scheduler parameter calculation module 335 can automatically update the AF values according to the type of event. According to an embodiment, the enhanced scheduler parameter calculation module 335 can also be configured to update the AF values in real-time to accommodate unforeseen events including changing weather patterns, natural or other disasters or law enforcement/military activity.
Application Factor with Dependency on Application Characteristics
According to an embodiment, the enhanced scheduler parameter calculation module 335 can be configured to extend the application factor (AF) from a function of application class and specific application to also depend on application characteristics. According to some embodiments, the AF is further adjusted based upon video frame size, video frame rate, video stream data rate, duration of the video stream, amount of data transferred with respect to the total amount of video stream data, video codec type, or a combination of any of these video application characteristics.
In an embodiment, the optimization criterion is to increase the number of satisfied users. Based on this criterion, the AF of a video data stream is adjusted by an amount inversely proportional to the data rate of the video stream. A lower AF may result in more packets being dropped during periods of congestion than would be dropped using a higher AF. For the similar amount of quality degradation, lowering the AF of a video stream of higher data rate may free up more network bandwidth than lowering the AF of a video stream of lower data rate. During the period of congestion, it is preferred to lower the AF of a video stream of higher data rate first, so the number of satisfied users can be maximized.
In an embodiment, the optimization criterion is to minimize perceivable video artifacts caused by imperfect packet transfer. Under this criterion, the AF of a video stream is adjusted by an amount proportional to the frame size, but inversely proportional to frame rate. For example, a lower AF may result in more frames being dropped during periods of congestion than would be dropped when using a higher AF. An individual frame of a video stream operating at 60 frames per second is a smaller percentage of the data over a given time period than an individual frame of a video stream operating at 30 frames per second. Since the loss of a frame in a video stream operating at 60 frames per second would be less noticeable than the loss of a frame in a video stream operating at 30 frames per second, the stream operating at 30 frames per second may be given a higher AF than the stream operating at 60 frames per second.
In an embodiment, the AF of a data stream may be adjusted dynamically by an amount proportional to the percentage of data remaining to be transferred. For example, a lower AF may be assigned to a data stream if the data transfer is just started. For another example, a higher AF may be assigned to a data stream if the transfer of entire data stream is about to complete.
In an embodiment, the AF of a video data stream is adjusted by a value dependent on the video codec type detected. A lower AF may be assigned to a video codec which is more robust to packet loss. For example, an SVC (H.264 Scalable Video Coding extension) video stream may be assigned a lower AF than a non-SVC H.264 video stream.
In an embodiment, the AF of a video data stream is adjusted based upon the duration of the video data stream, the amount of time remaining in the video data stream, or a combination thereof. For example, an operator may decide to assign a higher AF to a full-length Netflix movie as compared to a short 10 second Youtube clip, since the customer may have a higher expectation of quality for a feature length film as compared to a brief video clip. In another example, the operator may decide to dynamically assign a higher AF to a video data stream that is nearing completion as compared to one that is just starting in order to leave the customer who has finished viewing a video data stream with the best possible impression (see Recency Effect described below).
Information describing the duration of a video data stream may be obtained using the enhanced classification methods described above, including the Range information indicated during an RTSP message exchange. Information on the amount of time remaining in the video data stream may be calculated, for example, by subtracting the current video playback time from the stop time indicated in the Range information. Current video playback time may also be obtained by inspection of individual video frames or by maintaining a free-running clock which is reset at the beginning of playback. One skilled in the art would understand there may be alternate methods to obtain current video playback time.
In an embodiment, the AF of a video data stream is adjusted based upon the specific client device or device class used to display the video data stream. Device classes may include cell phones, smart phones, tablets, laptops, PCs, televisions, or other devices used to display a video data stream. Device classes may be further broken into subclasses to include specific capabilities. For example, a smart phone with WiFi capability may be treated differently than a smart phone without WiFi capability.
The specific device may refer to the manufacturer, model number, configuration, or some combination thereof. An Apple iPhone 4 (smart phone) or Motorola Xoom (tablet) are examples of a specific device.
The client device class, subclass, or specific device may be derived using various methods. In an embodiment, the device class may be derived using video frame size as described above. For example, the HTC Thunderbolt smart phone uses a screen resolution of 800 pixels by 480 pixels. The enhanced packet inspection module 410′ can detect or estimate this value using methods described above and determine the device class based upon a priori knowledge regarding the range of screen resolutions used by each device class or specific device.
In an embodiment, information regarding the device class, subclass or specific device is signaled between the client device and an entity in the network. For example, in a wireless network 100, a client device 150 may send information describing the vendor and model to the core network 102 when the client device initially joins the network. This information may be learned, for example, by the enhanced packet inspection module 410′ of a base station 110 for use at a later time.
Once learned, the device class, subclass, or specific device may be used to adjust the AF based upon operator settings. For example, in
In an embodiment, AF may be further modified by one or more service levels communicated via operator policy/SLA 350. For example, an operator may sell a mobile Netflix package in which customers pay additional fees in support of improved video experiences (e.g., quality, quantity, access) on their mobile phones. For customers participating in this program, the operator may assign an increased AF for the video stream application class shown in
In addition to selling retail services directly to the end user, a network operator may additionally or alternatively sell network capacity on a wholesale basis to a second operator (termed a virtual network operator or VNO) who may then sell retail services to the end user. For example, mobile network operator X may build and maintain a wireless network and decide to sell some portion of the network capacity to operator Y. Operator Y may then create a retail service offering to the general public which, possibly unbeknownst to the end user, uses operator X capacity to provide services.
In an embodiment, AF may be further modified by the existence of a VNO who may be using capacity on a network. For example, an operator X may have two VNO customers: Y and Z, each with differing service agreements. If operator X has agreed to provide VNO Y with better service than VNO Z, then data streams associated with VNO Y customers may be assigned a higher AF than streams associated with VNO Z customers, for a given device class, application class and specific application. In another example, operator X may sell retail services directly to end users and contract to sell services to VNO Y. In this case, the operator X may choose to provide its customers higher service levels by assigning a larger AF to streams associated with its customers as compared to those associated with VNO Y customers. Enhanced classification methods may be used to identify traffic associated with different VNO customers, including, for example, inspection of IP gateway addresses, VLAN IDs, MPLS tags or some combination thereof. One skilled in the art would recognize that other methods may exist to segregate traffic between VNO customers and the operator.
A further method to enhance the weight function extends the mapping coefficient, b, to a time varying function, assigned on a per queue basis. That is, b is a function of both time (t) and queue (q), b(q,t). In one embodiment, b(q,t) is adjusted in real-time, in response to, or in advance of, scheduler decisions for streams carrying video data streams (streaming or two-way) each on unique queues. This embodiment can further reduce peak load with minimal QoE loss by taking advantage of both the recency effect (RE) and duration neglect (DN) concepts as described by Aldridge et al. and Hands et al. See Aldridge, R.; Davidoff, J.; Ghanbari, M.; Hands, D.; Pearson, D., “Recency effect in the subjective assessment of digitally-coded television pictures,” Image Processing and its Applications, 1995, Fifth International Conference on, vol., no., pp. 336-339, 4-6 Jul. 1995, and Hands, D.S.; Avons, S. E., “Recency and duration neglect in subjective assessment of television picture quality,” Journal of Applied Cognitive Psychology, vol. 15, no. 6, pp. 639-657, 2001, which are hereby incorporated by reference.
The concept of DN is that the duration of an impairment viewed during video playback is less important than its severity. Thus for video being transported across a multiuser, capacity constrained network, it may be preferred (from a QoE perspective) for a scheduler which has already dropped one or more video packets from a video stream to continue to drop packets from that stream, rather than choose to drop packets from an alternate video stream, so long as the packet loss rate does not exceed a preset threshold. For example, based on the DN concept, discarding 5% of the packets of a single video stream over 10 seconds provides improved network QoE as compared to discarding 5% of the packets for 2 seconds, for each of 5 different video streams.
The concept of RE is that viewers of a video playback tend to forget video impairments after a certain amount of time and therefore judge video quality based on the most recent period of viewing. For example, a viewer may subjectively judge a video playback to be “poor” if the video had frozen (i.e. stopped playback) for a period of 2 seconds within the last 15 seconds of a video clip and judge playback to be “average” if the same 2 second impairment occurred 1 minute from the end of the video clip.
To this end, the coefficient ‘b’ of the enhanced weight equation (W′(q)=a*W(q)+b*AF(q)) or the enhanced credit equation (C′(q)=a*C(q)+b*AF(q)) is managed, on a per queue (and in this case a per data stream) basis, using the timing diagram shown in
As shown in
The method illustrated in
Once the entry condition or conditions have been met, a two-stage timing algorithm is initiated. A stream time is reset to zero (step 1120) and the value of b(i) is reduced by an amount Δ1 (step 1130).
A determination is then made whether the current frame discard rate exceeds a threshold for stream i (step 1140). For example, in an embodiment, the threshold is set to 5% over a 1 second period. In other embodiments, a different threshold can be set up for the stream based on the desired performance characteristics for that stream.
If the frame discard rate for the stream exceeds the threshold, the intentional degradation phase is terminated and the method continues with step 1155. Otherwise, if the frame discard rate does not exceed the threshold, a determination is made whether the timer has reached tdn. If the timer has reached or passed tdn, the intentional degradation phase is terminated and method continues with step 1155. Otherwise, if tdn has not been reached, the method returns to step 1140 where the determination is again made whether the current frame discard rate exceeds a threshold for stream i.
The coefficient b(i) is set to a value of b0+Δ2 (step 1155) before the timer is once again checked. A determination is then made whether the timer has reached tre (step 1160). If tre has not yet been reached, the method returns to step 1160. Otherwise, if the timer has reached tre, the method returns to step 1105.
According to an alternative embodiment, iteration through step 1160 can gradually adjust Δ2 towards zero over time period tre. According to another alternative embodiment, alternative (or additional) metrics such as packet latency, jitter, a predicted video quality score (such as VMOS) or some combination thereof is evaluated in step 1140. In a further embodiment, step 1140 is adjusted so that if the evaluation metric exceeds the threshold, the value Δ1 is reduced by an amount Δ3 with control then passing to step 1150 (rather than to step 1155).
In some systems, data identified as coming from two applications with different scheduling needs may be difficult to separate into separate queues for application of differing AFs, for example, for queues 491 and 491′ in
These problems can be overcome in various ways. In one embodiment, the data is split into separate queues 491 and 491′ which can be given different AFs. In this case, it is preferential to apply sequence numbers, ciphering, and header compression on the egress of the queues so that the data appears to have been pulled from a single queue with the scheduling order appearing to be the receipt order. This, however, is computationally complex and the order of processing, especially ciphering, may cause severe demand for computational resources. In another embodiment, rather than splitting queue 491 into queues 491 and 491′, the AF for queue 491 can be determined based on the combination of applications classes or specific applications currently carried on the data bearer rather than an individual application class or specific application. For example, if video data is detected on the logical link or bearer it may have an AF that is modified to reflect the QoE requirements of video even though the bearer may also have a background application that is periodically checking for email updates. When the use of video subsides, the AF may be returned to a value more appropriate for best effort data traffic. This is computationally less complex and achieves a similar result in cases such as streaming video when an application with demanding requirements is active most other data, if any, on the same bearer will be low in bandwidth relative to the demanding application. That is to say, the user will be concentrating on the video, voice, gaming, video conferencing, or other high bandwidth application while it is in use. To additionally guard against situations where the application with generally more demanding performance is not the bulk of the data on a bearer, for example playing a low bit rate YouTube video while email is downloading a very large attachment, the application factor can be a function of the percentage of traffic on the bearer from an application class or specific application rather than merely the presence of the application class or specific application.
The enhanced weight equation, W′(q)=a*W(q)+b*AF(q), and the enhanced credit equation, C′(q)=a*C(q)+b*AF(q), may be further modified to also include the effects of additional factors such as the current state of the queues, the current state of the communication link, and additional characteristics of the data streams. This may result in equations of the form:
W″(q)=a*W(q)+b*AF(q)+c1*F1(q)+c2*F2(q)+ . . . , and
C″(q)=a*C(q)+b*AF(q)+c1*F1(q)+c2*F2(q)+ . . . ,
where W″ is the modified weight and C″ is the modified credit, F1 and F2 are additional weight or credit factors, and c1 and c2 are coefficients for mapping the additional factors to the modified weight or the modified credit.
Adjusting the weights or credits using multiplicative additional factors rather than additive additional factors, or a combination of additive and multiplicative additional factors (e.g., W″(q)=a*W(q)+b*AF(q)*c1*F1(q)+c2*F2(q)+ . . . ) is possible, allowing scaling of weight or credit changes.
In an embodiment, a queue's weights or credits may be adjusted based upon queue depth. If a queue serving, for example, a video or VoIP stream reaches x % of its capacity, weights or credits may be dynamically increased by an additional factor until the queue falls below x % full, at which point the increase is no longer applied. The additional factor may be in itself application specific, for example with a different additional factor being applied for video than for voice, or may be dependent on the data rate of the service. In some embodiments, hysteresis is provided by including a delta between the buffer occupancy levels at which weight and credit increases begin and end. Additionally, when the queue is x′ % full, where x′>x, weights or credits may be further increased. In a further embodiment, a queue's weights or credits may be adjusted in part or in whole by a factor proportional to queue depth. These techniques allow additional factors to be applied to an individual stream in addition to or instead of an application factor (AF).
In another embodiment, a queue's weights or credits may be adjusted based upon packet discard rate. If a queue serving, for example, a video or VoIP stream exceeds capacity and packets are discarded, the discard rate is monitored. If the discard rate exceeds a threshold, weights or credits may be dynamically increased by an additional factor until the discard ceases or falls below the prescribed acceptable level, at which point the increase is no longer applied. The additional factor may be in itself application specific, for example with a different additional factor being applied for video than for voice, or may be dependent on the data rate of the service. In some embodiments, hysteresis is provided by including a delta between the discard rates at which weight and credit increases begin and end. Additionally, when the discard rate exceeds a higher threshold, weights or credits may be further increased. In a further embodiment, a queue's weights or credits may be adjusted in part or in whole by a factor proportional to packet discard rate.
In an embodiment, a queue's weights or credits may be adjusted based upon packet latency. If the average (or maximum over some time period) packet latency for a queue serving, for example, a video or VoIP stream exceeds a threshold, weights or credits may be dynamically increased by an additional factor until the packet latency falls below the prescribed acceptable level, at which point the increase is no longer applied. The additional factor may be in itself application specific, for example with a different additional factor being applied for video than for voice, or may be dependent on the data rate of the service. In some embodiments, hysteresis is provided by including a delta between the average (or maximum over some time period) packet latencies at which weight and credit increases begin and end. Additionally, when the packet latency exceeds a higher threshold, weights or credits may be further increased. In a further embodiment, a queue's weights or credits may be adjusted in part or in whole by a factor proportional to packet latency.
In an embodiment, a queue's weights or credits may be adjusted based upon packet egress rate. If the average (or minimum over some time period) egress rate for a queue serving, for example, a video or VoIP stream drops below a prescribed acceptable level, weights or credits may be dynamically increased by an additional factor until the egress rate rises above the prescribed acceptable level, at which point the increase in weights or credits is no longer applied. The additional factor may be in itself application specific, for example with a different additional factor being applied for video than for voice, or may be dependent on the data rate of the service. In some embodiments, hysteresis is provided by including a delta between the average (or minimum over some time period) egress rates at which weight and credit increases begin and end. Additionally, when the egress rate drops below an even lower threshold, weights or credits may be further increased. In a further embodiment, a queue's weights or credits may be adjusted in part or in whole by a factor inversely proportional to egress rate.
In rapidly changing RF environments, such as in a mobile network with adaptive modulation and coding, additional factors may be used to adjust the weights and credits rapidly based on airlink factors. When a user equipment has good receive signal quality for transmission from a base station, the base station, such as an LTE eNodeB, may transmit data to the user equipment at a higher data rate and/or with higher likelihood of successful reception. Likewise, when the base station has good receive quality for transmissions from the user equipment, the user equipment may transmit data to the base station at a higher data rate and/or with higher likelihood of successful reception. If the signal quality is observed to be highly variable, an additional factor can be applied to increase weights for a particular user equipment's data streams when the signal quality is good between the base station and that user equipment and decrease weights when the signal quality is poor, thereby providing the bandwidth to data streams for a second user equipment. The adjustment may be application specific. For example, the weight for a queue containing video may have an additional factor applied to ensure optimal use of good signal quality, while a delay and error tolerant service, such as email, for the same user equipment, may have a different or no additional factor applied, relying more on retries built into protocols such as TCP or the LTE protocol stack.
In addition to the additional factors that may be applied to weights or credits in response to the environmental factors described above, weights and credits or the application factors which modify them may be further modified based on knowledge of the transport protocols used. For example, a service that has one or more retry mechanisms available such as TCP retries, LTE acknowledged mode, automatic retry requests (ARQ), or hybrid-ARQ (HARQ) may have different additional factors applied for the life of the data stream or dynamically in response to such environmental factors as signal quality and discard rate or other indicators of congestion.
In an embodiment, the average bit rate of a data stream may be detected or estimated using techniques described above. Other methods may also be available depending upon the application. HTTP streaming, such as Microsoft HTTP smooth streaming, Apple HTTP Live Streaming, Adobe HTTP Dynamic Streaming, and MPEG/3GPP Dynamic Adaptive Streaming over HTTP (DASH), is one class of applications that supports video streaming of varying bit rate. In HTTP streaming, each video bitstream is generated as a collection of independently decodable movie fragments by the encoder. The video fragments belonging to bitstreams of different bit rates are aligned in playback time. The information about bitstreams, such as the average bit rate of each bitstream and the play time duration of each fragment, is sent to the video client (which may be a user equipment) at the beginning of a session in one or more files which are commonly referred to as playlist files or manifest files. This information may be detected by a network node such as a base station. In HTTP streaming of a live event, the playlist files or manifest files may be applicable to certain periods of the presentation, and the client needs to fetch new playlist files or manifest files to get updated information about the bitstreams and fragments in bitstreams.
Since the client has the information about bitstreams and fragments that it will play, it will fetch the fragments from bitstreams of different bit rates based on its current estimation of channel conditions. For example, due to variation in perceived channel conditions, a video client in a user equipment may fetch the first fragment from the bitstream of high bit rate, and the second fragment from the bitstream of low bit rate, and the next two fragments from the bitstream of medium bit rate. The channel conditions are often estimated by the video client based on information such as the time spent transporting the last fragment or multiple previous fragments and the size of these fragments. One deficiency of this approach is that the video client may not react fast enough to rapidly changing channel conditions. In one embodiment, the wireless access node, such as a base station, signals the current channel conditions to the video client, so the client can have more accurate information about the channel conditions and request the next fragment or the following fragments accordingly. In an alternative embodiment, the client may receive information regarding current channel conditions from the physical layer implementation, for example transmitter receiver module 279 of the station of
In one embodiment, the packet inspection module 410 (
Based on the dynamically calculated or estimated bit rate for a data stream, the weights or credits for a queue may be modified. In an embodiment, the dynamically calculated or estimated bit rate is compared to the queue egress rate and the queue's weights or credits are adjusted by the techniques described above. This may occur in response to detection of or absence of congestion. Additionally, in a case where a data stream was queued in a scheduling group scheduled by a weight based scheduling algorithm such as WFQ or WRR where weights were not based directly on bit rate, the data stream's queue may be moved to another scheduling group using a credit-based scheduling technique, such as PFS, basing credits on bit rates.
The packet inspection module 410 may compare the estimated bit rate of a specific application with the available channel bandwidth for transmission from the associated station. The instantaneous available bandwidth for transmission may be higher than the bit rate of the input traffic from a particular application. For example, an LTE base station using 20 MHz channels operating in 2×2 multiple-input, multiple-output (MIMO) mode has an instantaneous data rate of approximately 150 Mbps while a streaming video may have an average data rate of 2 Mbps and a peak data rate of 4 Mbps. In one embodiment, the wireless access node may buffer the data of an application and modify scheduler parameters to affect the instantaneous data rate and burst durations in advantageous ways.
Modifications of scheduler parameters may be combined to alter the outgoing traffic pattern 395 for the application to have packet transfer bursts that have high instantaneous bit rate and short duration relative to the incoming traffic pattern 390. This may have many benefits. If modulation and coding schemes are rapidly changing, for example due to mobility, the scheduler parameters may be modified to give preference to bursting the data at high rates during periods of good signal quality, effectively increasing the total system capacity through use of more efficient modulation and coding schemes for more of the data. It may also be desirable to increase the amount of idle time between two bursts, thereby making it possible to put the receiver at the user equipment into sleep mode for a longer time. This may be used to reduce the amount of time the user equipment receiver must be turned on to receive the data from the wireless access node. This can reduce the power consumption of the user equipment. This can be implemented, for example, to align with Discontinuous Reception (DRX) protocol in 3GPP HSDPA or LTE.
Those of skill in the art will appreciate that even though the above functions are generally described as if they reside in a station such as a base station, in some embodiments the functions may reside in other devices. Any device that performs queuing and scheduling may perform the algorithms. For example, a user equipment may perform the described algorithms when deciding how to schedule packets for uplink transmission or for deciding for which queues to request bandwidth uplink from the base station. A device or module that schedules bandwidth on the backhaul to or from a base station may perform the algorithms.
In one embodiment, the functions are distributed. For example, referring to
In an embodiment, information such as AF, alone or in combination with additional factors such as buffer occupancy, signal quality, discard rates, estimated bit rates, etc. may be used to compute an adjustment to the GBR setting typically established during the setup of a logical channel between network endpoints. The adjustment may be directed to mitigating congestion at the radio access network 550. For example, in an LTE network, an eNB scheduling parameter calculation module 335 may use the AF calculated for a particular data stream to request a modification of the corresponding data bearer's GBR by sending a message to the EPC packet gateway. In an alternate embodiment, an eNB scheduling parameter calculation module 335 may in addition request a QCI change, for example from a QCI which does not support GBR bearers to a QCI which does. Such requests may be made one or multiple times during the life of a data stream, and may be used alone or in combination with techniques described above, depending on conditions present at the eNB.
Processing of packets in the classification and queuing module 310 entails certain costs. For a function that is implemented in software running on a microprocessor, digital signal processor (DSP), or similar device, the processing cost is related to the computational complexity of the software instructions and resulting number of processor cycles (or instructions) and amount of random access memory (RAM) required to complete the processing. The number of processor cycles is often expressed in units of ‘millions of instructions per second’ (MIPS) or alternatively as a percentage of the total available MIPS for a given microprocessor (e.g., process X uses 50% of the total available MIPS). The amount of RAM is often expressed in units of ‘thousands of bytes’ (KB). For a function implemented in integrated circuit hardware, processing cost may be expressed in terms of the die area (e.g., square millimeters, number of gates, number of look-up-tables) used to perform this function and the power dissipation of the hardware (e.g., in milliwatts or watts). The processing cost can also be expressed in terms of increased solution cost and price to a customer. Therefore, efficient packet inspection is valuable to reduce processing cost.
Packets entering the packet inspection module 1500 via the bidirectional interfaces 1510, 1560 may be initially inspected by the traffic monitoring module 1520. The traffic monitoring module 1520 may inspect packets flowing in a single direction or both directions. In an embodiment, packets may be delayed in the packet inspection module 1500 via queues or buffers in order to provide time for other modules, for example, the connection detection module 1530 and the stream and session detection module 1540, to inspect and process packets identified for further inspection and processing. Alternatively, to limit transport latency, some or all packets (or portions of packets) may be copied for further inspection and processing while the original packets are forwarded to the next step in the path toward transmission. For example, the original packets may be supplied to the data queues 315 feeding the scheduler 330 in the parameterized scheduling module illustrated in
To improve packet processing efficiency, the packet inspection module 1500 may employ one or more techniques to filter packets based on simple criteria that have a low processing cost so that only a subset of the packets received by the packet inspection module 1500 undergo more complicated packet inspection that has a higher processing cost. Filtering the packets may also be viewed as selection of packets for further inspection.
In an embodiment, the traffic monitoring module 1520 may filter packets so that only uplink packets are inspected by the connection detection module 1530 or the stream and session detection module 1540. Filtering reduces the processing cost of detecting connections, streams, or sessions that are initiated by nodes at the edge of a network (for example, the user terminal device 560 of the wireless communication system in
For example, the traffic monitoring module 1520 may filter packets such that the connection detection module 1530 may receive and inspect only uplink packets to detect the initiation of a TCP connection via the detection of the SYN message sent from a client (e.g., user terminal device 560) to a server (e.g., data source 510). This technique may also be applied in reverse to improve processing efficiency for sessions initiated from a server (e.g., from the data source 510 or within the core network 102).
In an embodiment, one or more characteristics may be used to filter packets and reduce the processing cost to detect new connections based on protocols used. For example, knowledge that a mobile network operator (MNO) has configured its network using only a certain source IP address or source IP address range may be used when attempting to detect new UDP or TCP connections or streams. Alternatively, TCP source or destination port numbers may be used to filter packets. For example, to reduce processing cost an initial inspection stage may be employed to send only packets with headers containing TCP destination port 80 for further HTTP protocol processing.
In an embodiment, the traffic monitoring module 1520 may monitor only packets assigned to a specific class of service. For example, in an LTE radio access network, the traffic monitoring module 1520 may filter packets based on class of service and only pass packets corresponding to the lowest class of service, QCI=9, to the connection detection module 1530 and/or the stream and session detection module 1540 for further processing but ignore packets assigned to all other classes of service, QCI=1-8. In a further example, the traffic monitoring module 1520 may monitor all packets to or from users who have paid extra for an MNO's ‘Gold’ service level while packets to or from users participating in the MNO's ‘Silver’ or ‘Bronze’ service level may not be monitored. Many other filter systems are possible. Additionally, one or more filters may be employed in logical combination with each other and/or other detection techniques.
In an embodiment, filters based on packet size may be used in the traffic monitoring module 1520. For example, in detecting a particular packet message during either connection or session initiation, there may be a narrow range of packet sizes corresponding to the specific message of interest. A packet filter that only forwards packets for additional processing if the packets are within a size range (minimum and maximum) or above or below a size threshold may be used to reduce processing cost. For example, a video streaming session may be detected based on the characteristics of the real-time streaming protocol (RTSP). RTSP packets are encapsulated within TCP/IP frames and carried across an IP network, for example, as illustrated in the wireless communication system depicted in
RTSP establishes and controls multimedia streaming sessions with a client and a server exchanging the messages. A first RTSP message sent from the client to the server is a request message. The first line of a request message is a request line. The request line is formed with the following 3 elements: (1) Method; (2) Request-URI; and (3) RTSP-Version. RTSP defines methods including OPTIONS, DESCRIBE, ANNOUNCE, SETUP, PLAY, PAUSE, TEARDOWN, GET_PARAMETER, SET_PARAMETER, REDIRECT, and RECORD.
In an embodiment, the stream and session detection module 1540 may capture information during the DESCRIBE phase of the video streaming session setup by inspecting uplink packets identified for further processing by the traffic monitoring module 1520. A client DESCRIBE packet may be detected using a string (i.e., character text) match on the text ‘DESCRIBE’ contained in the RTSP message within the TCP payload. The server response in this case would be transported on the typically more heavily loaded downlink direction. This server response may contain critical information (e.g., an ‘m=video’ field as part of an SDP message which is the payload of an RTSP response message to an RTSP request message with DESCRIBE method) which may be used to determine the application class (e.g., video streaming). To reduce the processing cost to detect the server reply, the traffic monitoring module 1520 may be configured to only identify packets from the associated TCP connection for further RTSP processing if those packets have a payload size between 950 and 970 bytes. To further reduce processing cost, in an additional embodiment, the filtering of packets based on size and subsequent RTSP processing may only be active for a limited time duration or for a finite number of packets after detecting the DESCRIBE packet transmitted by the client. For example, a packet inspection system attempting to detect a DESCRIBE response, including the filtering technique above, may only be active for 1 second, after which the inspection process terminates.
In an alternative embodiment, the initiation of a video streaming session using the RTSP protocol may be detected by detecting an RTSP PLAY command issued from the client. The server response, typically carried to the client on the more heavily loaded downlink direction contains a playback range field that may be stored in the status module 1550. The detection of the RTSP PLAY response from the server may be improved, for example, by passing only packets of size 360-380 bytes for further RTSP processing. To further reduce processing cost, the filtering by packet size and RTSP processing may only be active for a limited time duration or for a finite number of packets after detecting the PLAY packet. For example, packet inspection to detect a PLAY response may only be active for 1 second, after which the inspection process terminates.
A packet or message size filter may be used to reduce the processing cost for other protocols, application classes, and specific applications. The traffic monitoring module 1520 may employ several filtering mechanisms simultaneously. For example, the traffic monitoring module 1520 may simultaneously filter by LTE bearer or QCI, filter on an already detected TCP connection, and filter on packet size for a finite time period.
The connection detection module 1530 inspects packets to determine when a network connection, used to support an application stream or session, has been initiated or terminated. The connection detection module 1530 may inspect packets identified for further processing by the traffic monitoring module 1520 to detect the initiation of a new TCP connection. Example connections may occur between the user terminal 560 and the data source 510 of the wireless communication system of
The connection detection module 1530 may also detect a connection by inspecting the packets in another connection. For example, in RTSP streaming, an RTSP request message with SETUP method, and the corresponding response message, which are transported in a TCP connection, include the information of the connection on which the video or audio packets will be transported. Below is an example of an RTSP request message with SETUP method sent from client “C” to server “S,” labeled with “C->S,” and the corresponding response message sent from server to client, labeled with “S->C.”
The RTSP request message indicates that the RTP packets and RTCP packets should be sent to the client at specific ports (4588 for RTP packets and 4589 for RTCP packets in the example). The response message echoes the client port information. In addition, it includes the server ports for the server to receive the RTP packets (6256 in the example) and RTCP packets (6257 in the example). Normally these two server ports are also used as source ports in packets sent from the server to the client. For this particular example, an RTP packet from the server to the client has source port number equal to 6256 and destination port number equal to 4588. An RTCP packet from the server to the client has source port number equal to 6257 and destination port number equal to 4589. An RTP packet from the client to the server has source port number equal to 4588 and destination port number equal to 6256. An RTCP packet from the client to the server has source port number equal to 4589 and destination port equal to 6257. After inspecting these two RTSP messages, the UDP connection for transporting RTP packets and the UDP connection for transporting RTCP packets can be detected.
In an embodiment, the traffic monitoring module 1520 may monitor packets in a unique manner (including the absence of monitoring) based upon the association of a packet with one or more of the following characteristics: logical link (e.g., LTE data bearer), connection (based on previous detection by the connection detection module 1530), data stream, application session (based on previous detection by the stream and session detection module 1540), class of service, network service level agreement (SLA), or network policy settings.
After a new connection has been detected by the connection detection module 1530, a context entry may be created in the status module 1550. After the detection of a terminated connection, a context entry may be deleted or modified in the status module 1550. In an embodiment, the status module 1550 maintains a context for each detected connection. The context may include characteristics for layers generally corresponding to a 7-layer networking model. Example characteristics include:
In an alternative embodiment, real-time or historical metrics may also be collected and stored in a connection's context entry. For example, a context entry may contain information regarding a connection's duration (e.g., seconds), number of bytes transferred, number of packets transferred, average bitrate (e.g., kbits/second), maximum bitrate (e.g., measured over a time interval). In addition to use in analytics, the real-time metrics may be used for reactive adjustment of scheduler parameters, such as application factors. The historical metrics may be used for predictive adjustment of scheduler parameters. A context may also contain session quality metrics (for example, packet loss statistics, packet retransmission statistics, and packet error rate) that may also be used to adjust scheduler parameters.
In an embodiment, the context stored in the status module 1550 may contain entries associated with active connections (i.e., those connections that have been initiated but not yet terminated). In an alternative embodiment, the context may additionally retain a history of connections including information regarding connections that have been terminated. In an embodiment, the context entries associated with terminated connections may contain the same information as entries for active connections (e.g., a combination of characteristics listed above). Alternatively or additionally, the context entries associated with terminated connections may contain information summarizing the connection history. For example, the context entry may contain a subset of the above characteristics plus information such as the total number of bytes transferred or the duration of the connection. In an embodiment, the context entries associated with active connections may inherit and carry the contexts of terminated connections when the active connections and terminated connections are related. For example, when a user fast forwards a YouTube video to a new starting point in the video, the current connection is terminated and a new connection is created. The context entry for the new connection can inherit the context of the terminated connection and retain the history and analytics information accumulated on the terminated connection.
In an embodiment, the context may be stored by the status module 1550 in the form of a file, array, linked list, or other suitable storage mechanism providing random read/write access.
Further packet inspection may be performed by the stream and session detection module 1540 to identify the initiation or termination of the streams comprising a session on a connection and to identify the application class, specific application, or other characteristics. Example characteristics that may be identified by the stream and session detection module 1540 include:
Many other connection, stream, session, and application characteristics could be identified in addition to or instead of those listed above.
In an embodiment, application class, specific application, and other characteristics described above, which have been detected by the stream and session detection module 1540, are added to a connection's context entry in the status module 1550.
The packet inspection module 1500 can be implemented in a single wireless or wireline network node, such as a base station, an LTE eNB, a UE, a terminal device, a network switch a network router, a gateway, a backhaul device, or other network node (e.g., the macro base station 110, pico station 130, enterprise femtocell 140, or enterprise gateway 103 shown in
In an embodiment, functions within the packet inspection module 1500 may be partitioned such that a subset of functions processes only data plane packets while a different subset of functions processes only control plane packets. For example, a function in the connection detection module 1530 used to detect a new UE or new data bearer in an LTE eNB base station may process only 3GPP control plane packets. Alternatively, a function in the connection detection module 1530 used to detect a new TCP connection on an LTE data bearer in an LTE eNB base station may process only data plane packets.
In step 1615, the connection detection module 1530 determines if the traffic monitored in step 1610 constitutes a new connection. In an embodiment, the connection detection module 1530 retains the state of the connection establishment protocol (e.g., TCP SYN, SYN-ACK, ACK messages) and identifies a new connection based upon a successful result from that protocol. In an alternate embodiment, the connection detection module 1530 compares the connection identification information gathered during step 1610 to the context stored in the status module 1550. If the connection identification information (e.g., logical link, IP addresses, UDP port numbers) matches an existing, active connection in the context stored by the status module 1550, then the connection information is deemed to be for an existing connection rather than a new connection and control returns to step 1610. However, if the connection information is not found in the existing, active context stored by the status module 1550, a new connection has been identified. At step 1620 the connection information is stored in the context stored by the status module 1550. The process then continues to step 1625 where monitoring of the connection is initiated for detection of information relating to the connection status and any streams, sessions, and applications associated with traffic transported on the connection. Then the process returns to step 1610 to monitor for new connections. The steps of the process for detecting initiation of connections may be performed concurrently. Additionally, the process may be modified by adding, omitting, reordering, or altering steps.
In step 1630, packets that are associated with the specific connection are monitored. Based on filtering criteria, the traffic monitoring module 1520, identifies packets related to the state of the specific connection for further processing by the connection detection module 1530 and identifies packets related to stream creation and termination and forwards those packets to the stream and session detection module 1540. The traffic monitoring module 1520 may also identify packets for further inspection for stream, session, or application information of interest. These packets may be forwarded to another module such as the other logic module 1570, the status module 1550, or the stream and session detection module 1540. For example, the traffic monitoring module 1520 may be configured to identify packets from a particular video stream periodically so that another module, for example, the other logic module 1570, may determine the current playback state. Alternatively or additionally, the traffic monitoring module 1520 may detect TCP retransmission requests for the particular connection so that the status module 1550 may record the metrics for use in assessing the quality of the service provided over the connection. The traffic monitoring module 1520 may also be configured to identify patterns in traffic and use the patterns to aid in application detection.
In step 1640, the connection detection module 1530 inspects packets to determine if the connection being monitored has been terminated. For example, for TCP connections, a FIN message pair with one message sent from each of the TCP server and the TCP client is the formal method of terminating a TCP connection. If a FIN message is detected from both TCP client and TCP server, then the connection detection module 1530 may conclude that the TCP connection has been terminated. To reduce computational complexity and processing cost, detection of only one or the other of the two FIN messages may be used to determine that a connection has been terminated. The processing cost may be further reduced when the connection detection module 1530 detects FIN messages only in the link direction that carries less traffic. For example, on many mobile networks, the uplink direction often carries less traffic than the downlink direction. Therefore, in this case detection of a FIN message on the uplink direction of link 190 requires fewer packets to be inspected and thus entails a lower processing cost than the detection of FIN messages on the downlink direction or the detection of both FIN messages. The termination of a TCP connection may also be detected by inspecting whether a packet has an RST flag set. Some sessions may have more than one connection. For example, an RTSP video streaming session has one TCP connection for transporting RTSP messages and multiple UDP connections for transporting RTP and RTCP packets. The UDP connections should be terminated when the TCP connection is terminated. In one embodiment, the termination of a connection is detected, if its associated connection is terminated.
Different methods for detection of initiation and termination of connections, streams, and sessions may have different costs, for example, in terms of processing power. The methods may also have different robustness. There could be a cost associated with a certain method whereby the method is only used if sufficient computational resources are available and a less robust but less costly method is used otherwise. Available computational resources could vary dynamically, for example, with temperature, battery charge level, power saving modes, or memory utilization. Computational resources may also vary as a function of network traffic load as measured by total system bitrate (e.g. megabits/second), packet rate (e.g. packets/second), number of active connections, streams, and/or sessions.
If the connection has been terminated as determined by step 1640, the status is updated in step 1650. In an embodiment, the entry and all information pertaining to the terminated connection may be removed from the context stored by the status module 1550. In an alternative embodiment, a historical record of the connection may be retained in the context entry along with an update of the entry's current status indicating that it is no longer active. This may be used for predictive updating of scheduler parameters. After updating the status module 1550, control proceeds to step 1655 where the process monitoring the connection is terminated. Termination of the process may include de-allocating resources used to perform the monitoring.
If the connection is not terminated, the process continues to step 1660. In step 1660, the stream and session detection module 1540 inspects packets to detect the initiation of a new stream or session and to identify the application class, specific application, or other session or stream characteristics. The detection of a new stream or session may cause the traffic monitoring module 1520 to modify the methods used to identify packets for further processing. For example, if the stream is determined to be a video stream over TCP, traffic monitoring module 1520 may be configured to periodically identify packets from which to detect or estimate video playback progress. The progress may be monitored, for example, by monitoring the TCP sequence number in an HTTP server's GET response and the client-side TCP ACK messages.
In an embodiment, previously detected characteristics (e.g., detected in step 1615 of the process for detecting initiation of connections of
For example, if traffic with a particular IP address yielded a reverse DNS lookup or WHOIS query that included the name ‘YouTube’ then this traffic stream could be considered a unidirectional video stream (Application Class) using the YouTube service (Specific Application). In an embodiment, a comprehensive mapping between domain names or assignees and application class and specific application can be maintained. The mapping may be periodically updated to ensure that the mapping remains up to date.
In an embodiment, the stream and session information detected in step 1660 in combination with the underlying connection information is compared to existing stream and connection information stored by the status module 1550. If a match to the detected stream and connection information is not found in the stored context, then the stream may be declared new and stored in step 1670 as a new stream entry associated with the underlying connection in the status module 1550.
In an embodiment, information about multiple streams may be compared to determine whether the new stream constitutes a new session or is part of an existing session. For example, if a stream is detected to be a video stream over RTP on the same logical link for the same user as a previously detected and still active voice stream over RTP and a previously detected recent SIP signaling stream, the combination of streams may be identified as a conversational video (e.g., video Skype) session.
In another example, voice over LTE (VoLTE) and interactive video combined with VoLTE may be detected. The detection may occur even though the IP Multimedia Subsystem (IMS) signaling of the setup of the services may be encrypted (as it is in LTE). For example, the packet inspection module 1500 may detect IMS signaling between the core network and a user equipment, followed shortly thereafter by the creation of a bearer or stream with a bit rate consistent with voice (e.g., 32 kbps). This information may be used to infer that a VoLTE session was initiated on the new bearer or stream. An example use of the information is by the scheduler parameter calculation module 335 of
In another example, if a stream is determined to carry streaming video with a certain playback range immediately following a stream that carried a portion of the same video with a different playback range, the two streams may be identified as part of the same video streaming session. Maintaining the status of the earlier stream (even after termination) by the status module 1550 allows this association to occur. In an embodiment, the saved context is updated with the stream, session, application class and specific application information described above. Such stream relationships may be used to determine device information. For example, detecting that multiple sequential streams versus a single stream are used for a YouTube video may be used to distinguish an Apple product using the iOS operating system from a device running the Android operating system. Detection of the stream, session, application, and device information may be used in the calculation of scheduler parameters such as application factors impacting weight and credits. The history may also be used for predictive modification of scheduler parameters.
Alternatively or additionally, detailed characteristics about the specific session may also be added to the context (step 1670 or step 1630) based on information available in packet headers or from packet stream profiling (as may be performed in step 1630). For example, the context describing a streaming video session may also include the following characteristics: video clip duration, resolution, frame rate, bit rate, container format, video coder-decoder (codec) format and configuration, client device (e.g., Android smart phone, Apple iPad, TV set-top box). The characteristics may be used, for example, to modify application factors used in scheduling. Other characteristics associated with streaming video, and with other application classes, may also be identified and stored in the context.
Once status or context has been updated in step 1670 or if a new session is not detected in step 1660, the process continues to step 1680. In step 1680, the stream and session detection module 1540 attempts to identify the termination of a stream and its associated session. As more than one stream may exist on a connection, in an embodiment, the process may attempt to identify the closure of more than one stream. Additionally, step 1680 may determine whether the termination of a stream constitutes termination of a session by comparing the stream to the context for the session. If the stream is the last active stream associated with a session, the session may be deemed terminated. Alternatively, a session may not be terminated immediately. For example, in the case of a session that is an instance of the YouTube application on an iPhone, the session may be made up of multiple sequential streams. Maintaining the session over these streams is beneficial in calculating scheduler parameters such that quality of experience is maintained.
Clients using the HTTP protocol to initiate a session may use an HTTP GET command to request an HTTP file with a specified content length from an HTTP server. In an embodiment, for sessions initiated using the HTTP protocol, session termination may be detected by monitoring the client-side TCP ACK number. If an HTTP server's GET response body starts with TCP sequence number N and the length of the HTTP response body (content length) is L, the session may be deemed terminated when the client sends a TCP segment with ACK number equal to N+L. Alternatively, to accommodate fixed bit length implementations, the session may be deemed terminated when a gap (for example, a minute or more) of no packets on a TCP connection follows a TCP segment with ACK number equal to (N+L) modulo 2 EXP B, where B is the bit length of the TCP segment number field, thus allowing the TCP sequence number to wrap around.
To reduce processing cost, the stream and session detection module 1540 may be configured to inspect the client ACK number periodically rather than continuously. Inspection for other information may also be performed intermittently over time. The intermittent processing may occur at regular or irregular time intervals. The inspection period may be fixed or may be adjusted based upon the number of packets remaining in a transmission. For example, after a new HTTP session has been detected, the stream and session detection module 1540 may monitor packets for 100 ms in each 1 second period. As the session nears completion, the stream and session detection module 1540 may be configured to inspect a larger percentage of packets as shown, for example, in the table below.
In the above example, session completeness may be calculated as current bytes transmitted (most recent client ACK number minus N) divided by the total bytes to be transmitted (L). Other techniques may be employed to adjust the packet monitoring period which may result in further improvements to processing cost and/or termination detection accuracy.
As there is risk that the detection of session termination is missed by employing the above technique, the stream and session detection module 1540 may also use this technique in conjunction with other methods such as session timeout (e.g., no session packets sent over a specified time period) or bitrate techniques, as described below.
If the termination of a session has not been detected, the process returns to step 1630. If in step 1680 it is determined that a session has been terminated, the process continues to step 1690 and the status is updated. In an embodiment, the status is updated by the removal of the current session, application class, specific application, and related information stored by the status module 1550. In an alternative embodiment, a historical record of the session may also be retained by the status module 1550. This historical record can include some or all of the session characteristics stored in the context while the session was active. Once the status has been updated, the process returns to step 1630 where further monitoring of the connection occurs. In an alternative embodiment for which only a single session may be associated with each connection, the process may proceed from step 1690 to step 1655.
In an embodiment, the steady state bit rate of a data stream may be used to identify the application class or specific application of a new session. For example, a session with a bidirectional data stream having a bitrate of 64 kbps may be characterized as a ‘voice’ application class, based on the bitrate associated with the G.711 codec. In an alternative embodiment, such a stream may be considered a voice application class only after the session has been ongoing for a time larger than a minimum time period (e.g., 3 seconds). To accommodate the proliferation of voice codecs with differing compression ratios and codecs with variable bit rate capabilities, the above technique may be further modified to detect bidirectional data streams with bitrates between a minimum and maximum value, such as 8 kbps to 64 kbps.
Similar techniques may be used to detect the presence of streaming video. For example, the packet inspection module 1500 may detect the presence of a high definition (e.g., 1080p) video streaming session by measuring that the average, unidirectional bitrate over a time period is within a predetermined minimum and maximum bitrate range (e.g., between 1 Mbps and 4 Mbps). In addition, the bitrate pattern (i.e. the bit rate measured and tracked over some time period) may also be used to determine the application class or specific application. For example, a YouTube video server using the HTTP protocol transmits data to an Android smart phone in a pattern of short, high rate bursts followed by long, very low rate quiet periods. An example of such a pattern is illustrated in the bitrate versus time graph of
The use of bitrates and/or bitrate patterns may be extended to detect other application classes or specific applications. Other examples include gaming, machine-to-machine communication, and video conferencing.
Additionally or alternatively, the use of bitrates and bitrate patterns may be used by the stream and session detection module 1540 to determine that a stream has been terminated (step 1680). For example, if a stream has been detected and is classified as a streaming video session (via bitrate detection or other methods), the stream and session detection module 1540 may measure the average bitrate (e.g., 2 Mbps) at the beginning of the stream and on a periodic basis thereafter. If the bitrate falls below a specified threshold (e.g., 10% of the measured average bitrate) over a specified period of time (e.g., 3 seconds) or across a specified number of samples (e.g., three 100 millisecond samples taken every second), then the stream may be deemed terminated. To reduce processing cost, the bitrate monitoring may be configured to be less frequent. Alternatively, to improve detection speed, the bitrate monitoring may be configured to be more frequent.
In an embodiment, the bitrate monitoring may be used or configured uniquely per stream or session. For example, for an HTTP based video streaming session, the termination scenarios may be considered to be of finite number and reliable. In such a scenario, bitrate monitoring may be used as a fallback or safety net to detect the unlikely cases of termination via unknown or unpredicted causes or in case the expected termination protocol is missed. In such a case, bitrate monitoring may be set to be very infrequent (e.g., every 10 seconds) to minimize processing cost. It may alternatively be disabled to minimize processing cost. In contrast, for sessions, streams, or connections having protocols, application classes, and/or specific applications unknown to the packet inspection module 1500, there is higher risk that the termination of the stream may not be detected based on the detection and inspection of protocol messages. In such a case, bitrate monitoring may be configured on a very frequent basis (e.g., every 100 milliseconds) since bitrate monitoring may likely be the only mechanism for detecting the stream or session termination.
In an alternative embodiment, the use of bitrate and bitrate patterns may be used by the connection detection module 1530 (step 1640) to determine that a connection has been left in an inactive and/or error state and should be deemed terminated. For example, if the average bitrate of a TCP based connection falls to zero over a specified length of time (e.g., minutes or hours), then the connection detection module 1530 may conclude that the connection has been broken in a manner that has not resulted in an orderly connection tear-down, for example, using FIN messages. In an alternative embodiment, the connection detection module 1530 may count TCP segments in one or both network directions. If the total number of segments is zero over a specified length of time, the connection detection module 1530 may conclude that the connection may be deemed terminated.
In an embodiment, application class or specific application may be established by inspection of the protocols that establish the session. For example, to identify HTTP based video streaming, the stream and session detection module 1540 may be configured to inspect the ‘Content Type’ field in a Hyper Text Transport Protocol (HTTP) packet. The content type field contains information regarding the type of payload based on the definitions specified in the Multipurpose Internet Mail Extensions (MIME) format as defined by the Internet Engineering Task Force (IETF). For example, the following MIME formats would indicate either a unicast or broadcast video packet stream: video/mp4, video/quicktime, video/x-ms-wm. To reduce processing cost, the application detection module may be configured to inspect packets for the ‘Content Type’ field in the downlink direction only after the successful detection of an HTTP ‘Get’ request in the uplink direction and only for a limited period of time (e.g., 2 seconds).
According to an embodiment, the stream and session detection module 1540 is configured to inspect the Host field contained in an HTTP header. The Host field typically contains domain or assignee information, which can be used to map the stream to a particular application class or specific application. For example, an HTTP header field of “v11.1scache4.c.youtube.com” could be inspected and mapped to Application Class=video stream, Specific Application=YouTube. In order to reduce processing cost for the detection of client initiated video sessions (for example, initiated by the user terminal 560 of the wireless communication system of
To further improve efficiency, in an embodiment, the above techniques may be logically combined so that the detection of the application class or specific application using one technique suspends additional packet inspection of the same connection by other techniques. For example, if one technique to detect YouTube is successful then packet inspection using the HTTP MIME approach may be suspended.
In an alternative embodiment, to further improve efficiency, the application class or specific application may be determined by the use of class of service (CoS) packet markings. For example, a MNO may decide to use LTE QCI=3 for real-time gaming and QCI=5 for IMS signaling and configure the packet inspection module 1500 in an LTE eNB with this information. Thus, packets arriving at the eNB with these characteristics may be quickly evaluated and removed from further processing.
In an embodiment, the termination of a logical link or messages relating to the termination of a logical link may be used by the connection detection module 1530 to determine that a connection has been terminated. For example, in an LTE network, signaling messages passed to the radio resource control (RRC) layer from the physical (PHY) layer indicating the loss of an RF link to a UE may be used by the connection detection module 1530 to terminate all sessions and connections associated with the UE.
In an embodiment, control plane messages carried across a network are used to detect the termination of a data plane connection by the connection detection module 1530. For example, access stratum (AS) control plane messages are sent by an LTE UE to a serving eNB to initiate and confirm handover of the UE to a new, target eNB. These messages may be detected by the connection detection module 1530 and may be used to declare the termination of all sessions, streams, and connections associated with the UE. In an alternative example, AS control plane messages between the eNB and UE are used for releasing (terminating) a dedicated data bearer. These messages may be detected by the connection detection module 1530 and used to declare that all connections associated with the data bearer have been terminated.
Congestion occurs when demand exceeds capacity. Congestion may occur at a number of domains, or levels within a communication system. One domain of congestion is the physical domain. The physical domain can have sub-domains, for example, addressing physical channel capacity or where in the network the congestion exists. The physical domain of congestion may, for example, address congestion of channel capacity of an entire communication channel, composite of all uplink and downlink communications, between a base station and multiple subscriber stations. For example, in the communication system of
Another domain of congestion is the policy domain of congestion. The policy domain can also have sub-domains. Policy domain congestion can occur when demand for bandwidth exceeds a policy limit. For instance, a group of services (e.g., members of a scheduling group or the services provided by a virtual network operator (VNO)) may be limited by operator policy to a subset of the bandwidth of the communication channel. In such a case, the group of services may experience congestion when its aggregate demand exceeds its allotted portion of the communication channel even if the communication channel as a whole is not congested. Additionally, an individual subscriber station may have restrictions on the amount of bandwidth it may use, either by policy (e.g., a limitation of its service plan) or by physical capabilities that restrict the subscriber station's peak data rates. A subscriber station may experience congestion due to these limitations even though the communication channel as a whole is not congested. Similarly, the subscriber station may experience congestion even if none of its services are members of groups experiencing congestion.
Other domains of congestion may also exist. The domains of congestion are not mutually exclusive. Additionally, interaction between domains may occur. Accordingly, a response to congestion may consider multiple domains. A communication network with devices that effectively detect and respond to congestion can manage the impact of congestion on QoE.
Congestion may be detected in various ways. Additionally, various devices may detect congestion. For example, a base station (e.g., the macro base station 110, pico station 130, enterprise femtocell 140, or residential femtocell 240 shown in
One method for detecting congestion determines whether demand exceeds a capacity threshold. The demand may be, for example, a measured demand, an estimated demand, or predicted demand. The capacity threshold may be, for example, a communication channel capacity or a percentage of a capacity. Whether demand exceeds a capacity threshold may be a simple ‘greater than’ comparison. Whether demand exceeds a capacity threshold may also be more complex, for example, including temporal factors or a combination of parameters.
Comparing a metric to a threshold can take numerous forms. In one embodiment, a metric is compared to a threshold and if the threshold is exceeded, an action is taken. There may be one threshold for indicating a congestion event or quality impacting event has occurred and another that indicates the condition has cleared. In another embodiment a metric is compared against a set of thresholds, for instance indicating a variety of severities of congestion, and the action taken is dependent upon which threshold is crossed. In a further embodiment, a metric may represent a continuous range of severities of a condition, such as congestion, and may be mapped to a continuous range of actions, for instance a multiplicative factor applied to a scheduler parameter.
Another method for detecting congestion uses its impact on communication resources. Example resource impacts include packet delay or latency and scheduler buffer queue depth or occupancy.
Congestion may also be detected from its impact on performance of associated communication devices. Examples of performance impacts include dropping packets due to scheduler buffer overflow, dropping packets due to aging out of packets, and an ingress data rate for a stream that is greater than its egress rate. Additionally, congestion may be detected using protocol metrics, for example, protocol delays, retransmissions, or packet loss in protocols such as UDP, TCP, or HTTP.
Another method for detecting congestion uses a two-step (or multi-step) process. A simple (but less accurate) measurement can be made to detect possible congestion and trigger an accurate (but more complex) measurement to detect actual congestion. For example, a simple higher layer protocol measurement exceeding a threshold can trigger the use of a more complex metric.
The detection of congestion may be further used to measure or predict the effects of congestion on QoE. The effect on QoE may be for streams for particular application classes or specific applications. Predicted effects on QoE can be used, alternatively or additionally with congestion measurements, in initiating control responses to adjust scheduling, for example, to adjust an application factor applied to scheduler weights or credits for the stream or other streams competing for the resources.
Measuring whether demand exceeds capacity may be accomplished using a number of methods. For example, bandwidth demand in the form of input traffic 305 ingress bit rates into the classification and queuing module 310 in the parameterized scheduling system of
Another example of detecting whether demand exceeds capacity is to measure physical resource usage and compare that usage to a threshold that, if exceeded, indicates or predicts congestion. For example, a metric such as “Total PRB usage” may be used to measure physical resource block (PRB) usage in LTE systems (see 3GPP TS 36.314 V10.2.0, titled “3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Evolved Universal Terrestrial Radio Access (E-UTRA); Layer 2—Measurements (Release 10)”). A related metric, which may be used to measure congestion for a subset of services, also defined in 3GPP TS 36.314, is “PRB usage per traffic class” which measures PRB usage by groups of services in the same QCI. Such metrics may be calculated by, for instance, the scheduler module 320 of
Measuring the effects of congestion on resource or communication performance may be accomplished using a number of methods. Measuring the effects of congestion may create metrics for packet delay or latency, packet discard, the difference between packet arrival rates or times and packet delivery rates or times, or a combination, thereof. For example, when a packet is received by a station, the packet may be placed in a queue or buffer prior to being scheduled for transmission to a user device. The time between receipt by the station and transmission to the user device is the latency or delay of the packet through the station. Packet delay metrics may be measured for a communication link as a whole, individual logical links or services, groups of services, individual devices, or groups of devices, for example, the group of devices serviced by a VNO or class of service. 3GPP TS 36.314 defines such a metric, “Packet Delay in the DL per QCI.” This metric may be further averaged over all QCIs to determine the average delay for the communication link as a whole and variants may be constructed for individual user equipment or services. When a delay metric exceeds a threshold, it can be an indication of congestion, an indication of changed QoE, or both.
Metrics measuring the initial delay of services or applications may also be used to indicate congestion or an impact to QoE. For instance, the portion of call setup time delay due to congestion for services initiated with the SIP or Real Time Streaming Protocol (RTSP) protocols may provide a metric for congestion or QoE created by measuring the difference between the receipt time of the initial protocol packet and its transmission across the communication channel. The initial protocol packet may be detected, for example, by the packet inspection module 410 of
Congestion may cause packets to be discarded and affect QoE. Discards due to congestion may occur because of buffer overflow. When the buffer space allocated to a scheduler queue or set of queues is exhausted, there is no place to store a newly received packet. Either the new packet must be discarded or a previously received packet may be discarded. Measurement of discards due to buffer or queue overflow exceeding a rate threshold may be used to detect congestion and estimate the impact on QoE. Additionally, the scheduler buffer occupancy or depth may be measured. As the scheduler buffer occupancy increases, the likelihood of a packet discard due to buffer overflow increases. Accordingly, scheduler buffer occupancy exceeding a threshold may be used as an indication of congestion that is predicted to impact QoE in the near future. In addition to discards due to buffer overflow, in many systems packets may be discarded because they have been buffered longer than a predetermined time limit. Discard due to aging of packets exceeding a threshold may be used as a metric for congestion. 3GPP TS.314 describes such a metric, “Packet Discard Rate in the DL per QCI.” This metric may be further averaged over all QCI to determine the average discard rate for the communication link as a whole and variants may be constructed for individual user equipment or services
Relative packet movement rates may also be used as a metric for congestion. For example, if packets for a service, user device, class of service, or system are being received with an ingress rate greater than the transmit egress rate, congestion may be occurring or about to occur. For example, using the parameterized scheduling system 300 of
Measurements on higher layer protocols may also be used to detect congestion. For example, TCP protocol measurements may be performed by the packet inspection module 1500 of
Messages in the HTTP protocol may be detected using methods similar to those described above. The time difference a station detects between an HTTP “get” on the UL and the corresponding HTTP response on the DL can be used to indicate congestion somewhere in the total round trip path between a server somewhere in the Internet and a client on a user device excluding the link between the station and the user device. This metric may be used in conjunction with TCP metrics to determine whether congestion is on the communication link between the station and the user devices 150 or elsewhere, such as in the Internet.
Those of skill will appreciate that the various illustrative logical blocks, modules, controllers, units, and algorithm steps described in connection with the embodiments disclosed herein can often be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, units, blocks, modules, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular system and design constraints imposed on the overall system. Skilled persons can implement the described functionality in varying ways for each particular system, but such implementation decisions should not be interpreted as causing a departure from the scope of the invention. In addition, the grouping of functions within a unit, module, block or step is for ease of description. Specific functions or steps can be moved from one unit, module or block without departing from the invention.
The various illustrative logical blocks, units, steps and modules described in connection with the embodiments disclosed herein can be implemented or performed with a processor, such as a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general-purpose processor can be a microprocessor, but in the alternative, the processor can be any processor, controller, or microcontroller. A processor can also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.
The steps of a method or algorithm and the processes of a block or module described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module (or unit) executed by a processor, or in a combination of the two. A software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of machine or computer readable storage medium. An exemplary storage medium can be coupled to the processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium can be integral to the processor. The processor and the storage medium can reside in an ASIC.
The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles described herein can be applied to other embodiments without departing from the spirit or scope of the invention. Thus, it is to be understood that the description and drawings presented herein represent a presently preferred embodiment of the invention and are therefore representative of the subject matter, which is broadly contemplated by the present invention. It is further understood that the scope of the present invention fully encompasses other embodiments that may become obvious to those skilled in the art.
This application is a continuation-in-part of U.S. patent application Ser. No. 13/549,106, filed Jul. 13, 2012, which is a continuation-in-part of U.S. patent application Ser. No. 13/396,503, filed Feb. 14, 2012, which is a continuation-in-part of U.S. patent application Ser. No. 13/236,308, filed Sep. 19, 2011, which is a continuation-in-part of U.S. patent application Ser. No. 13/166,660, filed Jun. 22, 2011, which are hereby incorporated by reference. This application is also a continuation-in-part of international patent application No. PCT/US12/43888, filed Jun. 22, 2012, which is hereby incorporated by reference. U.S. patent application Ser. No. 13/166,660 is a continuation-in-part of U.S. patent application Ser. No. 13/155,102, filed Jun. 7, 2011, which claims the benefit of U.S. provisional patent application Ser. No. 61/421,510, filed Dec. 9, 2010, which are hereby incorporated by reference. U.S. patent application Ser. No. 13/166,660 is also a continuation-in-part of U.S. patent application Ser. No. 12/813,856, filed Jun. 11, 2010, now U.S. Pat. No. 8,068,440, which claims the benefit of U.S. provisional patent application Ser. No. 61/186,707, filed Jun. 12, 2009, U.S. provisional patent application Ser. No. 61/187,113, filed Jun. 15, 2009, and U.S. provisional patent application Ser. No. 61/187,118, filed Jun. 15, 2009, which are hereby incorporated by reference.
| Number | Date | Country | |
|---|---|---|---|
| 61421510 | Dec 2010 | US | |
| 61186707 | Jun 2009 | US | |
| 61187113 | Jun 2009 | US | |
| 61187118 | Jun 2009 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 13549106 | Jul 2012 | US |
| Child | 13607559 | US | |
| Parent | 13396503 | Feb 2012 | US |
| Child | 13549106 | US | |
| Parent | 13236308 | Sep 2011 | US |
| Child | 13396503 | US | |
| Parent | 13166660 | Jun 2011 | US |
| Child | 13236308 | US | |
| Parent | PCT/US12/43888 | Jun 2012 | US |
| Child | 13166660 | US | |
| Parent | 13155102 | Jun 2011 | US |
| Child | 13166660 | US | |
| Parent | 12813856 | Jun 2010 | US |
| Child | 13166660 | US |
