The invention relates to methods and apparatus for improving communications in digital networks. The invention also relates to grouping of bandwidth allocations and burst groups in digital networks.
Traffic shaping is important in digital networks. Traffic shaping involves buffering traffic and sending traffic based upon a desired profile. A traffic profile can include, but is not limited to, the following properties: a level of priority relative to other traffic, buffer depth, latency through the buffer, jitter in sending the traffic contained in the buffer, and a rate at which the traffic should be sent. A common approach to traffic shaping involves the use of a queuing system to manage the profile. As traffic arrives, it is placed on the queue. The traffic is de-queued based upon its assigned drain rate.
In certain situations it may be necessary to restrict a group of queues to a predefined amount of overall bandwidth. Doing so creates burst groups, in which the member queues compete for a common resource (bandwidth), but do not affect others outside the group. This allows the network to be better managed, where physical network connections can be subdivided into virtual “pipes” or “connections”.
Problems with some prior devices include, for example, lack of scalability, sheer size and high gate-count cost per queue for decentralized shaping engines, expensive caching/arbitration mechanisms, and lack of ability to shape traffic with fine granularity across a broad spectrum of desired rates, or groups of rates.
Preferred embodiments of the invention are described below with reference to the following accompanying drawings.
This disclosure of the invention is submitted in furtherance of the constitutional purposes of the U.S. Patent Laws “to promote the progress of science and useful arts” (Article 1, Section 8).
Attention is directed to a commonly assigned patent application Ser. No. 10/224,508, titled “System and Method for Shaping Traffic from a Plurality Of Data Streams Using Hierarchical Queuing,” and naming as inventors Keith Michael Bly and C Stuart Johnson, which is incorporated herein by reference. Attention is also directed to a commonly assigned patent application Ser. No. 10/224,353, titled Bandwidth Allocation Systems and Methods, and naming as inventors Keith Michael Bly and C Stuart Johnson, which is incorporated herein by reference.
When there are a large number of profiles or services to manage (e.g, more than 32), it is desirable to group or aggregate like profiles or services together to compete for common resources. This is desirable, for example, in order to protect one “type” of traffic from another, where “type” is a broad term used to classify traffic based on the needs of the moment. For example, a type of traffic could be “video traffic,” “pay-per-view” video traffic, “all traffic for customer X,” all email traffic, all traffic with a given priority, all traffic with the same MAC-DA (same first 6 octets of a frame), etc. This allows prevention of bursty traffic, for example, from stealing bandwidth from very smooth, jitter-intolerant traffic.
It is desired to smooth the streams as shown by curves 28, 30, and 32 in
The solution provided in accordance with one embodiment of the invention, based on the above commonly assigned patent applications, is to utilize multiple credit sources (burst groups), and to assign each queue 44-47 (
The shaping engine 34 can be defined, for example by a microprocessor, or other digital circuitry. The burst group manager 12 includes linked lists 40 (see
Pointers and linked lists are known in the computer arts. A pointer is a variable that points to another variable by holding a memory address. A pointer does not hold a value but instead holds the address of another variable. A pointer points to the other variable by holding a copy of the other variable's address. A read/write pointer keeps track of a position within a file from which data can be read or written to. A linked list is a chain of records called nodes. Each node has at least two members, one of which points to the next item or node in the list. The first node is the head, and the last node is the tail. Pointers are used to arrange items in a linked list, as illustrated in
More particularly,
The shaping engine 34 (see
The queues 44-47 can have shaping profiles, which include properties such as: priority, depth, latency, jitter, and rate. For example, video needs to always get through. A large amount of latency is not desirable for video, as any latency will cause the resulting picture to become jerky, and fall behind. The same is true of the rate at which video is sent. A constant, consistent stream should be used to supply the video information “just in time” for the next entry or element (e.g., packet or frame) of the picture on a TV or computer. Therefore, “video” traffic is properly classified so that it is managed appropriately. Because the video must always get through, it is given a “high” priority. Because video cannot be influenced/slowed-down with a large amount of latency, the depth of the queue is selected to be shallow. Therefore, little data can build up, waiting in the queue. With regard to rate, the video queue gets its own bandwidth end-to-end on a switch, and does not have to compete with any other queue for bandwidth. Queues for other classifications of traffic would similarly have appropriately chosen priorities, depths, latencies, jitter, and rates.
In the illustrated embodiment, the rate-algorithm for the shaping queues 44-47 is a centralized time division multiplexing algorithm that is implemented, for example, by the shaping engine 34. More particularly, in the illustrated embodiment, the rate-algorithm for shaping traffic across many queues uses a table based credit allocation scheme. A fixed size bandwidth allocation table (BAT) 50 is traversed at a constant rate. Each location (e.g. row) 68-75 (
Queue Rate=(total credit in table for this queue)÷(time to traverse table)
As long as there is enough traffic to keep the queue from being empty, this drain rate can be maintained indefinitely. The rate itself is calculated by dividing the amount of credit listed in the table 50 by the time it takes to traverse the table 50 one time. A queue 44-47 is considered eligible to send an entry or element (e.g., a packet or, more particularly, a frame) when the queue 44-47 has acquired enough credit to send the entry in question.
In the illustrated embodiment, the shaping engine 34 manages both adding and deleting from the shaping queues, as well as updating the shaping queues with bandwidth tokens from the bandwidth allocation table 50.
Based upon the needs of the design in which this queuing structure is implemented, the size of the table 50 can be adjusted to provide the desired minimum and maximum achievable rates. The minimum rate is defined by one credit divided by the table traversal time, and the maximum rate is defined by the maximum number of entries allowed in the table, each containing the maximum number of credits, divided by the table traversal time. The maximum number of entries allowed in the table 50 is dictated by the implementation. For example, the maximum number of entries allowed in the table, is determined by the overall “profile” of the port(s) 18 supported by this queuing structure, etc. More particularly, the maximum number of entries allowed in the table is determined by the circuitry or software that manages traversing the table 50 relative to the number of queues 44-47 in the implementation, and how it manages updating the credit for each queue 44-47. Though a certain number of queues is shown in
As the bandwidth allocation table 50 is traversed, the queue listed in the entry 68-75 requests the credit listed from its assigned burst group or groups. The burst group or groups respond with whatever credit they currently have available, if any. Over time, as long as the burst group or groups in question are not oversubscribed with queues requesting more credit than is available, the queues all get the credit they request. However, if a burst group is oversubscribed, not all queues will receive all the credit they request from it, thus protecting the overall system credit from “greedy” groups of queues. Only queues are listed in the bandwidth allocation table 50; burst groups earn credit in a different manner.
In one embodiment, burst groups earn credit more often than the queues, but in relatively lower amounts each time they are updated. This is intentional; and results in the burst group's credit being made more available across the entire time it takes to traverse the bandwidth allocation table 50. This results in a better distribution of credit across the bandwidth allocation table 50, allowing for more options when configuring the bandwidth allocation table 50. This burst group update rate is represented by an “Nth” request interval between burst group credit updates in
More particularly, in one embodiment, burst groups earn credit using a simple periodic add function such as the following:
The queues 44-47 have an upper bound on the amount of credit they can accumulate. This protects the system 10 by not allowing a queue that has been idle for some time to suddenly saturate the system with newly arrived traffic. A beneficial side-effect of this limit is that, over time, as the queues which are located earlier than others in this “Nth” request interval no longer need credit (due to, for example, a lack of traffic), the queues listed later can gain access to more of the burst group's credit. This creates a natural order of precedence, which can be taken advantage of when configuring the bandwidth allocation table relative to the burst group update interval. This creates the ability to differentiate queue types (e.g., high precedence versus best effort queues). This is a dynamic assignment, in that a given queue can be configured either way (high precedence versus best effort) and changed on the fly by adjusting the configuration of the bandwidth allocation table 50 while traffic is running.
This can be extended further by intentionally sequencing queues in the BAT such that a queue that may have made a request early in the burst group interval (early in the bandwidth allocation table 50) is listed again at the end of the interval where it can request a maximum request value. This is shown in
While
In step 82, an entry 68-75 is read from the bandwidth allocation table 50.
In step 84, the amount of credit listed in the read entry 68-75 is requested from the credit allocation circuit or mechanism 51 of the queue's assigned burst group or groups.
In step 86, credit is added from the burst group's response to the queue's credit bucket.
In step 88, a determination is made as to whether the queue has enough credit to send a frame. If so, the frame is sent in step 90.
In step 92, a determination is made as to whether this entry is the last entry in the bandwidth allocation table 50. If so, the BAT index is reset to the beginning 68 of the bandwidth allocation table 50 in step 94. If not, the BAT index is incremented in step 96 to the next location or row in the bandwidth allocation table 50.
In step 98, a determination is made as to whether this is the Nth request for credit from the burst groups. If so, credit is updated for all burst groups in step 100 and process flow continues at step 82. If not, process flow skips step 100 and continues at step 82.
The preferred embodiment provides a solution that is scalable, and provides the ability to shape traffic for a variety of implementations in a cost effective manner. This results in a smaller overall design.
The preferred embodiment of the invention provides a centralized queuing structure, capable of supporting one or more ports, with a high queue density count. This centralized queuing structure is capable of dynamically supporting different ports over time, rather than a fixed set of queues only able to support a single port or ports. The design of the preferred embodiment is also scalable. The design of the preferred embodiment, by its very nature, can be implemented for one queue up to the feasible limits of today's technology, without significantly increasing the size of the central engine. The only increase to the cost of increasing size is the space needed for the linked-list management. Further, the design of the preferred embodiment by its very nature can be implemented to support an infinite variety of min./max. rate relationships. Previous implementations could only perform gross granularity transitions for various desired rates.
The preferred environment is all of Ethernet. Slight modification to “shaping” profiles would allow for use in any communications technology including, for example, ATM and SONET.
In one embodiment, the first queuing stage is included in a single ASIC, which provides for sufficient clock-speed to support Gigabit Ethernet rates.
Various alternative embodiments are possible. For example, one alternative embodiment has a reduced or increased number of queues.
In compliance with the statute, the invention has been described in language more or less specific as to structural and methodical features. It is to be understood, however, that the invention is not limited to the specific features shown and described, since the means herein disclosed comprise preferred forms of putting the invention into effect. The invention is, therefore, claimed in any of its forms or modifications within the proper scope of the appended claims appropriately interpreted in accordance with the doctrine of equivalents.
Number | Name | Date | Kind |
---|---|---|---|
5164938 | Jurkevich et al. | Nov 1992 | A |
5748629 | Caldara et al. | May 1998 | A |
5872769 | Caldara et al. | Feb 1999 | A |
5953318 | Nattkemper et al. | Sep 1999 | A |
5999518 | Nattkemper et al. | Dec 1999 | A |
5999563 | Polley et al. | Dec 1999 | A |
6031573 | MacCormack et al. | Feb 2000 | A |
6052375 | Bass et al. | Apr 2000 | A |
6067298 | Shinohara | May 2000 | A |
6084856 | Simmons et al. | Jul 2000 | A |
6157955 | Narad et al. | Dec 2000 | A |
6195355 | Demizu | Feb 2001 | B1 |
6259699 | Opalka et al. | Jul 2001 | B1 |
6275497 | Varma et al. | Aug 2001 | B1 |
6343081 | Blanc et al. | Jan 2002 | B1 |
6438134 | Chow et al. | Aug 2002 | B1 |
6477144 | Morris et al. | Nov 2002 | B1 |
6487212 | Erimli et al. | Nov 2002 | B1 |
6628652 | Chrin et al. | Sep 2003 | B1 |
6714553 | Poole et al. | Mar 2004 | B1 |
6754206 | Nattkemper et al. | Jun 2004 | B1 |
6950400 | Tran et al. | Sep 2005 | B1 |
6980552 | Belz et al. | Dec 2005 | B1 |
7042841 | Abdelilah et al. | May 2006 | B2 |
7058789 | Henderson et al. | Jun 2006 | B2 |
7072295 | Benson et al. | Jul 2006 | B1 |
20010001608 | Parruck et al. | May 2001 | A1 |
20010009552 | Parruck et al. | Jul 2001 | A1 |
20010017866 | Takada et al. | Aug 2001 | A1 |
20010018711 | Morris | Aug 2001 | A1 |
20010030956 | Chillariga et al. | Oct 2001 | A1 |
20010038628 | Ofek et al. | Nov 2001 | A1 |
20010055303 | Horton et al. | Dec 2001 | A1 |
20010055319 | Quigley et al. | Dec 2001 | A1 |
20020023168 | Bass et al. | Feb 2002 | A1 |
20020034162 | Brinkerhoff et al. | Mar 2002 | A1 |
20020044567 | Voit et al. | Apr 2002 | A1 |
20020071387 | Horiguchi et al. | Jun 2002 | A1 |
20020191622 | Zdan | Dec 2002 | A1 |
20030076848 | Bremler-Barr et al. | Apr 2003 | A1 |
20060233156 | Sugai et al. | Oct 2006 | A1 |
Number | Date | Country | |
---|---|---|---|
20040042399 A1 | Mar 2004 | US |