Patent Application
20050052996
The invention relates to the field of communications systems and more specifically to the management and admission control of voice over IP (VoIP).
Telecommunications networks and other networks are increasing in both size and complexity in order to serve the growing demand for high-speed communication networks for the transfer of voice and/or data information. As these telecommunication networks grow, alternate solutions or networks are sought to meet the demand for increasing network bandwidth and new IP-based services.
Traditionally, voice calls are transported entirely over the end-to-end, circuit-based Public Switched Telephone Network (PSTN). However, considerable attention has been directed toward the implementation of real-time communication across computer data networks, and particularly the ability to route voice traffic to and from the PSTN. Interest has also been raised in using Voice over IP (VoIP) solutions to facilitate voice communication between originating and terminating PSTN end points via an IP network. Using the Internet for long haul routing substantially bypasses the PSTN.
For PSTN bypassing applications, voice traffic is processed into IP (or ATM) packets, transported over an IP network (or ATM network), and then processed back to PCM voice. To facilitate such call routing, originating and terminating End Office (EO) switches can be connected to PSTN/IP (or PSTN/ATM) gateways that reside as hosts on the IP (or ATM) network. Based on the called number or other signaling indicator, the EO switches route certain calls through the IP (or ATM) gateways instead of the PSTN.
Unfortunately, when a new VoIP telephone voice call is established (with the intent of it being routed between two gateways in the network), there are no means to evaluate the level of congestion of the core IP network. In other words, it is possible to have too many new voice calls being introduced to the network at the same time so that the core IP network is overloaded. Under such a condition, it is highly likely that packets of information that contain the voice data will either be lost or delayed from arriving at the destination gateway. Both of such conditions result in poor Quality of Service (QoS) of the network which is an undesirable.
Two possible solutions to ascertaining the level of congestion of the core IP network is to overprovision the network such that packet loss will not occur and to reserve sufficient bandwidth between gateways and count voice calls on each path. However, the first solution requires expensive overbuilding of the network and the second solution is relatively complex to operate. Accordingly an improved means for establishing an admission policy of voice calls to a network is desirable.
The disadvantages heretofore associated with the prior art are overcome by a novel method and apparatus for analyzing the level of voice call traffic in an IP network before allowing a new voice call to enter the IP network.
In particular, in one embodiment, analysis circuitry is provided to each gateway between a PSTN and the IP network. Such circuitry obtains information from the IP network regarding the level of voice call traffic being transmitted from one gateway to another gateway. A parameter is calculated by the analysis circuitry based on obtained information. This parameter is then compared to predetermined thresholds to guarantee acceptable quality for a new voice call that is attempting to enter the IP network. If the value of the parameter is below a lower threshold, voice call quality is highly acceptable and the new voice call is allowed into the IP network. If the value of the parameter is between the lower threshold and an upper threshold, voice call quality is acceptable, but the new voice is allowed into the IP network at a reduced bandwidth in comparison to existing calls in the network whose parameter was below the lower threshold. If the value of the parameter is above the upper threshold, voice call quality will be unacceptable and the new voice is not allowed into the IP network.
In one embodiment, the parameter is a packet loss ratio (PLR). The PLR considers lost, late and received packets measured at a particular gateway along a particular path and reported back to a gateway that had sent such packets. The PLR is calculated by the equation A/(A+B) where A is the sum of lost and late packets arriving at the particular gateway along the particular path and B is the total number of successfully received packets arriving at the particular gateway along the particular path.
In particular, the apparatus includes a gateway for interfacing voice call data from a PSTN to an IP network. The gateway further includes a first circuit for passing said voice call data to the IP network, a second circuit for polling the IP network about traffic information transmitted therein and a third circuit for processing the polled information to determine whether the voice call data is to be accepted by the IP network.
In one embodiment, the first circuit includes one or more interface cards that are connected to the IP network and the second circuit is at least one strongarm card connected to said interface card via a host CPU circuit. The third circuit compares the parameter (PLR) based on the polled information to the upper and lower thresholds to make the appropriate decision of allowing, blocking or allowing at reduced bandwidth a new voice call. In doing so, quality of all calls on the network is maintained and new calls are not permitted into the network until such time that their quality is at minimum requirements.
The teachings of the present invention can be readily understood by considering the following detailed description in conjunction with the accompanying drawings, in which:
To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the figures.
The subject invention establishes and manages VoIP traffic in a network (for example an Internet Protocol (IP) network) by monitoring certain criteria indicative of network capability and instantaneous load. Accordingly, an exemplary telecommunications system is described as one potential environment in which a subject invention operates and exists.
It is also possible to bypass the PSTN 110 using the data network 118. Such an alternate communication path is established by connecting the first end office 106 to a first gateway 114 and likewise connecting the second end office unit 108 to a second gateway 116. First gateway 114 and second gateway 116 may be a single unit (as shown as a single structure 114 or may be represented by one or more independent structures A,B and C of second gateway 116. First and second gateways 114 and 116 respectively reside as hosts on the network 118. They provide VoIP services on behalf of the first wire line subscriber 102 and second wire line subscriber 104 and other users (not shown) communicating over the network 118. During VoIP communications between the first wire line subscriber 102 and the second wire line subscriber 104, PCM traffic is routed from the first end office 106 and second end office 108 to the respective gateways 114 and 116 for routing across the data network 118. Call control is managed through the Softswitch 112. When a new call is set up or a completed call is torn down, signaling messages are exchanged between the first end office 106 and the Softswitch 112 and between the Softswitch 112 and the second end office 108. The Softswitch 112 also acts as a gateway controller and exchanges messages with each gateway. In some networks there may be different Softswitches 112 controlling each gateway(114, 116 or the like) and these Softswitches exchange signaling messages with each other.
Within each gateway (114 and 116) is a corresponding admission control module (first admission control module 2041 is in first gateway 114 and second admission control module 2042 is in second gateway 116). These admission control modules 204x monitor VoIP traffic and generate the necessary reports to decide which calls will be granted access through the network 118 to maintain overall quality of service for all subscribers. A Measurement-Based Call Admission Control (MBCAC) algorithm contained within said admission control modules 204x is described in greater detail below.
The MBCAC algorithm operates in each gateway (114,116) independent of the other gateways in the network. Call quality statistics for a Real Time Transport Protocol (RTP) stream reflect the congestion status of the path followed by that stream. Thus, by observing these statistics, one can decide on the congestion status of the network paths.
Packets traveling to a destination gateway can follow different paths based on the port 206x chosen for the specific RTP flow. The MBCAC algorithm assumes that the selection of a port 206x for an incoming call request is under the control of a call controller in the gateway. Hence, the MBCAC algorithm keeps separate admission policies for the paths from different ports to the same destination gateway. It is also assumed that multiple calls going from a particular port to the same destination gateway follows the same path, i.e., there is no load balancing within the network other than provided by the gateways through the selection of an egress port. This assumption can be satisfied if the gateways use the system IP address of the destination gateway as opposed to the IP addresses of its ports. In this framework, load balancing is supported by controlling the egress port at the source gateway (i.e., first gateway 114). Since each egress port would map into a unique path in the IP network 118, the load from source gateway 114 to a destination gateway (i.e., second gateway 116) can be partitioned into different paths, resulting in load sharing in the network.
The destination gateway 116 receives the RTP packets generated by the source gateway 114 (e.g., at port E2) and addressed to itself. For each RTP stream, the receiver measures call quality statistics like packet loss ratio, delay and interarrival jitter for the stream. The measured statistics are sent back to the source gateway 114 periodically in a special field within the RTP packets or in RTCP packets. In one example, these statistics reflect the network conditions for the path following (E2-ER1-Network-ER3-E4). Thus, the MBCAC algorithm utilizes the call quality statistics of this flow to derive the congestion status of the directed path, uniquely defined by the source gateway E2, destination gateway pair.
The call quality statistics sent by the destination gateway 116 are collected by an RTP termination point in this gateway and formed into a call quality report. To support the MBCAC function, RTP flows are grouped into sets represented by (Egress port, Remote Gateway)-pair, i.e., there is a list of RTP flows for each (Egress port, Remote Gateway)-pair. Using the example introduced in
The flow diagram begins at step 402 with the soft switch 112 receiving a call set of requests from the PSTN network 110 (as per
At step 406, the soft switch 112 sends a message to second gateway 116 that includes the IP address and RTP port of first gateway 114 upon which the call is being set up as well as information about the destination PSTN switch. Since second gateway 116 now has information about first gateway 114, second gateway 116 checks the admission control policy of the path to first gateway 114. If the path is congested, an error message is generated at step 408 indicating that there is insufficient bandwidth to establish the desired path. If the call is to be accepted (i.e., there is sufficient bandwidth available to set up the call), second gateway 116 creates an RTP port (one of the egress ports 206x) and maps this into a voice trunk to the destination PSTN switch. This information is sent back to soft switch 112 as an “add response” message at step 410. This message is forwarded by the soft switch 112 to first gateway 114 so that the RTP port in first gateway 114 can be modified to include a transmit direction. At step 412 the necessary modifications are made to the TDM trunk based on the response from second gateway 116.
Next, first gateway 114 consults with the admission control algorithm to see if there is a path to the second gateway 116 that is not congested. If first gateway 114 is unable to find an uncongested path to second gateway 116 it sends an error message at step 414 to the soft switch 112. In such a scenario where the call attempt has been denied the call set up process is terminated by “subtract command” messages sent by the soft switch 112 to the first and second gateways, 114 and 116 respectively at step 416. In response to the subtract command messages of step 416, first gateway 114 and second gateway 116 provide subtraction command responses to the soft switch 112 at step 422 thereby completing the denied call set up attempt. If there is sufficient bandwidth available to set up the incoming call based on the first gateway 114 admission control algorithm results, a “modify response” message is sent back to the soft switch 112 at step 418. This signals the soft switch 112 that the data path for the voice call is ready for data transmission in both directions. Resultantly, an RTP session is established at step 420 and the voice call begins.
For the purposes of the subject invention, late packets are defined as packets that are discarded at the destination gateway (for example second gateway 116) since they are too late to be played or otherwise incorporated into the active voice call. Additionally, it should be noted that lost, late and received packets (collectively “sent” packets) are defined for the outgoing direction of a voice call, measured by the destination gateway (second gateway 116) and reported back to a source gateway (for example first gateway 114) in the opposite direction. Since each direction of the call takes a separate path through the IP network, there is a separate admission control decision for each direction of the call. All packet counts are defined per RTP connection. Furthermore, packet counts used in the packet loss ratio computation are counts measured over the most recent reporting period. In one embodiment of the invention, the reporting period is approximately two seconds; however, one skilled in the art will realize that various other reporting periods are possible dependent upon hardware and software being used in the overall system and network as long as the desired results are achieved. Other quality information could involve delay or delay variation.
The method continues at step 808 where a first admission policy is established. In one embodiment of the invention, the admission policy consists of two threshold values. In the first decision step 808, the first threshold value (a lower threshold P_low is introduced. The computed PLR is compared to the lower threshold P_low. If the PLR is less than P_low, the method proceeds to step 810 where the policy is set to accept new calls without any limitations. The method then awaits the next reporting period and loops back to step 804 to obtain the new quality of service information to continue evaluation. A reporting period is in the range of approximately 5-60 seconds and in one embodiment is 5 seconds. The exception reporting option will make this updating faster during congestion.
Should the PLR be higher than the lower threshold, the method proceeds to step 812 where the PLR is compared to a second threshold (a high threshold P_high). If the PLR is larger than the lower threshold P_low, but lower than the higher threshold P_high, the method proceeds to step 814 where the policy is set to admit new calls at reduced bandwidth. Such action reduces the bandwidth of new incoming calls to an extent that still allows quality of service. Similar to the accepted call scenario, accepted-bandwidth limited call step 814 loops back to step 804 to await the next reporting period to attain quality of service information.
Should the PLR be higher than the high threshold, the method proceeds to step 816 where the policy is set to block all or some percentage of new call requests from entering the network. In other words, path congestion has reached such a limit that an unacceptable number of packets are either being lost or received too late to be part of a call. As such, it is realized that no new calls can enter the network and maintain an acceptable quality of service level; therefore, such calls are not allowed into the network until path congestion is sufficiently reduced and quality of service can be maintained for all subscribers. The method ends at step 818.
Bandwidth reduction (as discussed above in step 814 of method 800) is achieved in a few different methods. One method is to physically change the encoder that is being used for the particular voice call. That is, there may be two or more encoders in a gateway (114 or 116) that carry out encoding tasks (one encoder having high bit rate characteristics and another having lower bit rate characteristics). If a bandwidth-reduced call is accepted, the encoder with the lower bit rate characteristics is used. When conditions allow for non-bandwidth limited channels, the system can switch back to the higher bit rate encoder. Another way in which bandwidth can be reduced is to use the same codec but increase the packet size. This will reduce the relative packet overhead. Another way of reducing bandwidth as per step 814 is to reduce the bitrate by activating silence suppression for the voice call. Briefly, silence suppression results in reducing bandwidth requirements since no packets are sent during silence periods. In most conversations only one person is talking so that, on average, in any one direction there is speech to send at most half the time. Thus suppressing packets representing silence can save considerable bandwidth. Note that silence suppression does not apply to fax calls, where picking a very large packet size would be more useful.
The above calculations were given in terms of packet loss ratios. The computation of packet loss ratios involve a division operation, which can be avoided if a loss and late packet count is used. In this case, P_low and P_high are converted to low and high packet count thresholds using the packet generation rate of the flow. It is assumed that the sum of received, lost, and late packets will be fixed, and equal to the number of packets transmitted by the local gateway in RTPQOS reporting period. For example, if a codec for a RTP flow is to generate 50 packets per second, there will be 100 packets transmitted every 2-second interval. Using this assumption, it is possible to use the number of lost and late packets instead of packet loss ratio. Thus, it is possible to define packet loss ratio thresholds in terms of packet count thresholds. Continuing in the example, the lower threshold would be (lost+late)_low=100*p_low, and the higher threshold would be (lost+late) high=100*p_high. This way, the SARM 310x can decide if an exception report (a block-call message explained in greater detail below) should be generated or not without performing any division operation. A variation to this would be to use a computed value for the sum of lost and late packets such that this sum is equal to the “Number of packets sent by the local gateway in the RTPQOS reporting interval-local.received”. This way, the inaccuracies related to loss packet estimation in the remote gateway are avoided.
Different RTP flows would have different packet rates; hence, the SARM 310x should take the packet rate of each flow into account. For example, with a 2-second measurement interval 1% packet loss ratio corresponds to only 1 lost packet if the packet rate is 50, while it corresponds to 2 lost packets if the packet rate is 100. The threshold values in terms of number of packets per flow will be provided by the call control during the call set-up. Moreover, if a flow is using silence suppression, the number of packets sent by the local gateway (114 in the above example) should be adjusted to reflect the silence suppression. Quantization of the packets may cause inaccuracies. As such, SARM 310x compares the value of local.received with the expected value, and if there is a large difference, it sets a flag or uses fraction and computes the packet loss ratio. Another technique omits the first set of measurement values for a newly set-up call. This way, the effect of network delay on the expected local.received is avoided.
The SARMs 310x send blocking rules to the admission control module 2041 in the shelf controller 302 as a result of the analysis conducted by method 800. To reduce the amount of transmitted data, the blocking rules may be reported as exceptions. If the determined admission rule is Accept, nothing is reported to admission control module 2041. However, if the rule is Reduce or Block, the SARM 310x reports the value to the admission control module 2041 through the host CPU 312 as an exception report. Once an exception report (Reduce or Block) is sent to the admission control module 2041 for a flow, there should be no reporting for the same RTP flow for a time interval of length T_u, which is called “exception update interval”. This update interval could be different than the periodic update interval if periodic updates are used instead of exception reports. One exception to this rule is if the last reported rule is Reduce and the newly computed rule is Block, the new value should be passed to the admission control module 2041 immediately. (Note that this is not applicable when Reduce rule is disabled by setting P_low to zero.) In this case, there should be no more reporting for the same RTP flow for a time interval equal to the update interval. A new exception report should be generated if the Reduce or Block rule is determined using a QoS report that was received after the update interval is over. This type of periodic exception reporting should be continued until an Accept rule is detected for the RTP flow. There is no need to report an Accept rule.
An alternative to the exception reporting per RTP is to perform exception reporting per (local interface, remote IP address). This way, the number of update messages can be greatly reduced. This approach results in a maximum of two reports generated within an update interval per (local interface, remote IP address)-pair as opposed to being per RTP flow.
The exception report, delivered to the admission control module 2041, includes the routeID of the flow that the measurement belongs to and the blocking rule. The admission control module 2041 uses routeID to determine the local interface and remote IP address of the flow. Note that the information regarding the mapping between the routeID and the (local interface, remote IP address)-pair should be located in the shelf controller 302. If this is not possible, the host CPU 312 should provide the explicit information as local interface and remote RTP address when submitting an exception report to the admission control module 2041.
When the admission control module 204, is initialized, the rules database 306 is empty. With time, blocking rules will be reported by the SARMs 310x. This blocking information is kept in the rules database 306 which is used by the admission policy function. An entry in this database is indexed by the Remote IP address. Moreover, each entry consists of subentries. Each subentry contains a blocking rule, an index to a local Ethernet interface, and a timestamp for the subentry. This way, each subentry shows the admission rule for the path defined by (local interface, Remote IP address)—pair. The number of subentries for Remote IP address is variable, where the maximum number is equal to the number of local interface cards, configured in the system. Note that the rules database 306 reflects the congestion status of the network paths from the local gateway to remote gateways. The opposite direction is handled similarly in the remote gateway. When the admission control module 2041 is initialized, information about the existing interfaces is determined.
The admission control module 2041 continuously listens to the exception reports from the host CPUs 312. When an exception report is received for an RTP flow, the blocking rule for the endpoints of the flow is updated. A method of updating the blocking rules is shown in
Should the IP address of the remote gateway be found at step 906, the method proceeds to step 910 where a second decision step is invoked. The second decision step 910 determines if a subentry for the local egress port 208x is found. If such subentry is found, the method proceeds to step 914 where the admission policy is updated and the timestamp is reset to the current time. If the subentry is not found, the method proceeds to step 912 where an appropriate subentry is created and the timestamp is set to the current time. At the conclusion of either of steps 912 or 914, the method proceeds to its final step at 916.
Admission control module 2041 periodically revises its database to remove subentries that were not updated in the last T_u+Delta seconds. The interval T_u is the time window where the SARMs 310x suppress reporting packet loss values for a connection following a report for the same connection. Here, Delta should be slightly larger than the QoS reporting period of the gateways, so that SARM 310 will have a chance of sending a second exception report before the admission control module 2041 removes the blocking rule. Delta can be 3 seconds since the RTP QoS reporting is done every 2 seconds. Based on this scheme, the admission control module 2041 assumes that a path is not congested if there is no exception report within the most recent time interval of length last T_u+Delta seconds. This action is beneficial because the host CPU 312 does not report when the packet loss ratio goes below the threshold value. Reporting of the below threshold value crossing is not needed since there might be more than one flow responsible for the blocking rule, which should be relaxed only if there is no update for the rule in the last T_u+Delta time interval by none of the related flows. Since the mapping of a rule for a path to all the flows, which reported exceptions for the path, would be computationally inefficient, an indirect method is used to remove the blocking condition. The period to revise the database to detect aged blocking rules should be chosen as small as possible to keep the database size small so that search operations will be efficient during a call to function as explained below.
Although various embodiments which incorporate the teachings of the present invention have been shown and described in detail herein, those skilled in the art can readily devise many other varied embodiments that still incorporate these teachings.
