WO2012065477A1

WO2012065477A1 - Method and system for avoiding message congestion

Info

Publication number: WO2012065477A1
Application number: PCT/CN2011/080049
Authority: WO
Inventors: 唐保; 徐春松; 王然; 吕勇; 刘奇峰
Original assignee: 中兴通讯股份有限公司
Priority date: 2010-11-18
Filing date: 2011-09-22
Publication date: 2012-05-24
Also published as: CN101984608A

Abstract

A system for avoiding message congestion includes: a package processor configured to: be connected to a traffic manager, classify, parse, modify and encapsulate a message, and carry out congestion processing according to a congestion strategy received from the traffic manager; and the traffic manager is configured to: carry out congestion detection on the message, generate the congestion strategy and send the same to the package processor. Also disclosed is a corresponding system. The present invention, on the one hand, reduces the memory overhead and periphery access overhead caused by maintaining enormous queues by the package processor; on the other hand, it saves enormous resource consumption by the traffic manager, improves the message processing performance and efficiency, and improves the reliability and stability of the system by way of the feedback mechanism.

Description

Message congestion avoiding method and system

Technical field

The present invention relates to the field of data communications, and in particular, to a packet congestion avoiding method and system. Background technique

With the rapid development of the network, the rapid growth of technology and performance of network equipment, the IP services carried on the Internet (the network) are gradually diversified, and the rich data, voice, video and other integrated services are booming. As network bandwidth increases from 10 Mbps, 100 Mbps, 1 Gbps, and 10 Gbps to 100 Gbps, it is becoming less and less sufficient for suppliers to increase bandwidth and expand network equipment. Based on end-to-end Quality of Service (QOS) management tools, it has gradually become the focus of attention of suppliers. Different types of services are provided for different customers and different services, such as: providing dedicated bandwidth, reducing packet loss rate, and reducing packet transmission delay and delay jitter. For this reason, technologies such as flow classification and coloring, traffic policing, traffic shaping, network congestion management, and network congestion avoidance are commonly used.

When the network is congested, all incoming packets are discarded when the length of the queue reaches the specified maximum length according to the traditional tail drop processing method. For a TCP packet, if a large number of packets are discarded, the Transmission Control Protocol (TCP) is globally synchronized. When multiple TCP connections enter the slow start at the same time, the traffic sent to the specified queue is too large. Frequent fluctuations in the ground, so that the flow on the line always fluctuates between very little and full. Congestion avoidance is such a technology. Its typical application is to supervise the use of network resources. When detecting the tendency of congestion, the discarding strategy is adopted to actively discard packets randomly. Since the packets are randomly discarded, multiple TCP connections are prevented from being simultaneously slowed down, thereby avoiding the global synchronization of TCP. When a TCP connection packet is discarded and decelerated, other TCP connections still have a higher transmission speed. In this way, whenever there is always a TCP connection for faster transmission, the utilization of the line bandwidth is improved, and the network is over-congested.

Congestion avoidance is performed on the basis of traffic classification, including: drop priority, congestion detection, congestion discard, and so on.

The main criteria for discarding priorities are: IP precedence, differentiated service code points (Differentiated Services Code Point (DSCP) values and MPLS EXP (MPLS: Multi-Protocol Label Switching, EXP: Experiment, tag experiment bit). The user can distinguish the different discarding priorities of the same flow queue, so that when the network is congested, the discarding probability of the difference is realized.

Currently, the main methods of congestion detection are: Tail Drop, RED, Random Early Detection, and Weighted Random Early Detection (WRED).

When the tail is discarded, all incoming packets will be discarded when the queue length reaches the specified maximum threshold.

When RED is used, the user can set the threshold of the queue. When the length of the queue is less than the low value, the packet is not discarded. When the length of the queue is between the low value and the high value, the random start is started. The packet is discarded. The longer the queue is, the higher the probability of discarding. When the length of the queue is greater than the threshold, all packets are discarded.

The difference between WRED and RED is that the former introduces IP precedence, DSCP value, and MPLS EXP to provide different discarding probabilities for packets of different priorities.

On current network equipment, congestion avoidance is mainly done by two components: packet processor (NP: Network Processer, etc.) and traffic manager TM (Traffic Manager)

There are two main implementations of existing congestion avoidance: one is implemented on the traffic manager; the other is done by the packet processor.

The packet processor performs the traffic classification, the traffic shaping, and the congestion avoidance according to the policy of the user. Congestion management and other operations; The other is to complete the packet classification, traffic policing, traffic shaping, congestion avoidance, congestion management and other operations directly by the packet processor.

The main problems with implementation one are: the packet processor completes the packet flow classification, the traffic manager performs congestion avoidance, and the traffic manager is located downstream of the packet processor. When congestion is discarded, packets are discarded in the traffic manager, which is a waste of resources compared to packets being discarded in the packet processor.

The main problems with implementation 2 are: Congestion avoidance is done directly on the packet processor. It is not only necessary to complete the traffic classification of the packet on the packet processor, but also to complete the congestion detection of the packet, and the congestion detection of the packet means that a large number of queues need to be implemented and maintained in the software of the packet processor, and Discard policy processing. For packet processors, implementing and maintaining a large number of queues introduces additional memory overhead, multiple access to peripherals, and performance and efficiency services are not guaranteed. Summary of the invention

The invention provides a packet congestion avoiding method and system, which improves message processing performance and efficiency.

To solve the above technical problem, the present invention provides a packet congestion avoidance system, including: a packet processor, configured to: connect with a traffic manager, complete packet classification, parsing, modification, and encapsulation; according to the received The congestion policy sent by the traffic manager performs congestion processing. The traffic manager is configured to: detect congestion of the packet, generate the congestion policy, and send the congestion policy to the packet processor.

Preferably, the packet processor includes: a message classification module, a classification policy library, and a discarding priority library, where the message classification module is configured to: query the classification policy library, the discarding priority library, and obtain Classification ID, drop priority.

Preferably, the packet processor further includes: a congestion avoidance module and a congestion policy library,

The congestion avoidance module is configured to: query the congestion policy database to obtain the congestion policy according to the received classification identifier and the drop priority sent by the packet classification module.

Preferably, the congestion policy library in the packet processor is configured to: receive the congestion policy that is sent by the traffic manager.

Preferably, the congestion avoidance module in the traffic manager is configured to: feed back the congestion policy to a congestion policy library in the packet processor.

Preferably, the congestion avoidance module in the traffic manager is configured to: generate a congestion avoidance policy according to a congestion detection result of the packet queue.

To solve the above technical problem, the present invention further provides a packet congestion avoidance method, including: the traffic manager generates congestion avoidance according to a congestion detection result of a packet queue and a drop priority. Policy and feedback to the packet processor;

The packet processor performs congestion processing according to the received congestion avoidance policy.

Preferably, the packet processor includes: a message classification module, a classification policy library, and a drop priority library,

The packet classification module queries the classification policy library, the discarding priority database, and obtains a classification identifier and a discarding priority.

The congestion avoidance module queries the congestion policy database according to the received classification identifier and the discarding priority sent by the packet classification module, and obtains the congestion policy.

Preferably, the congestion policy library in the packet processor receives the congestion policy fed back by the traffic manager; wherein a congestion avoidance module in the traffic manager feeds back the congestion policy to the packet processor Congestion policy library.

Preferably, the congestion avoidance module in the traffic manager generates a congestion avoidance policy according to the congestion detection result of the packet queue.

Compared with the prior art, the traffic manager congestion avoidance module generates a congestion avoidance policy according to the congestion detection result and the drop priority, the congestion avoidance policy includes a drop probability function, and the congestion avoidance in the packet processor is updated. The packet processing policy in the policy library, the packet processing policy includes: packet discarding/forwarding flag, and the packet processor and the traffic manager cooperate to complete the congestion avoidance of the packet. BRIEF abstract

The drawings are intended to provide a further understanding of the invention, and are intended to be illustrative of the invention. In the drawing:

1 is a schematic structural view of a congestion avoidance system of the present invention;

2 is a flow chart of the congestion avoidance system of the present invention. Preferred embodiment of the invention

The invention will be described in detail below with reference to the drawings in conjunction with the embodiments. It should be noted that the embodiments in the present application and the features in the embodiments may be combined with each other without conflict.

The system for congestion avoidance of the present invention includes:

a packet processor for classifying, parsing, modifying, encapsulating, and accessing various data packets to obtain related information;

The traffic manager, which performs congestion detection and marking of packets, and its performance should match the packet processor;

a classification strategy library, which is used to provide a packet classification policy to the packet processor message classification module, located on the storage device;

Drop the priority library to save the discarding priority configuration information after the packet is classified, and use it on the storage device.

A congestion avoidance policy library for providing supervisory policy and status information to the congestion avoidance module in the packet processor is located on the storage device.

The system and implementation method of the congestion avoidance of the present invention will be described below with reference to the accompanying drawings.

1 is a schematic structural diagram of a congestion avoidance system of the present invention, and a packet processor (for example, a network processor) of the present invention includes:

The traffic interface module 101 includes multiple physical interfaces to connect to different networks, and is a data channel for network packets to enter and exit the chip.

In the data receiving direction, the data frame received from a physical interface is removed from the encapsulation data packet of the communication interface, and is stored in the inbound and out data packet buffer together with the status information of the data packet; in the data transmission direction, according to the inbound and outbound data. The status information in the packet cache, the data packet is fetched, and the communication interface encapsulation is issued at the corresponding physical interface. A physical interface refers to a specific data communication interface entity, including an Ethernet interface, a POS interface, or other interface that can be used to exchange data.

The message classification module 102 sorts according to the classification policy library 103.

The user-configured classification policy library 103 is the basis for the block classification. According to the user-configured policy, with the Mac address, inner and outer 802.1p, inner and outer Vlan ID, MPLS Exp, IP Tos, 5-tuple, port number, QPPB GroupID, etc., a unique correspondence of the message can be obtained. Logo, It can also be called a stream ID, which is FlowID. When the congestion is sent, the WRED discarding of the queue packets is supported. The WRED processing is based not only on the flow identifier but also on the discarding priority (DP) of the packet. To do this, the user also configures a WRED drop priority library 104. Based on this configuration information, this module can find the discard priority (DP) information that can get two bits.

The congestion avoidance module 105 performs congestion avoidance according to the classification result of the packet classification module 102, that is, the flow identifier (FlowID) and the discard priority level (DP) in the discarding priority pool 104.

The congestion avoidance module 105 queries the congestion policy library 106 according to the information such as the flow identifier (FlowID) and the discarding priority (DP) in the classification result of the packet classification module 102, and returns the policy information (for example: packet discarding/forwarding) Marking, discarding probability), according to the policy information (message drop/forward flag, or drop probability), choose to perform packet forwarding and discarding actions.

The traffic interface module 107, the same traffic interface module 101, is only used on the message sending side.

The traffic manager (or TM) of the present invention includes:

Flow Interface Module 108. The main internal sub-modules are: 1. Receiver controller: The function is to receive bus data, including: signal conversion, detecting user-side data, and filtering data that does not meet the requirements, such as error packets. The data packet is analyzed, and the high priority protocol packet and the general data packet are distinguished and sent to the high priority protocol packet through the first in first out queue (FIFO) and the general receiving interface FIFO. The flow control information of the bus is collected, and the upper module is flow-controlled according to the state of the receiving interface FIFO. Second, the interface first-in-first-out queue FIFO, the role is to perform data buffering of the input interface. The interface FIFO must be able to receive data larger than the maximum packet length. When a complete packet is stored in the FIFO, an indication is sent to the traffic policing module.

Traffic policing module (Policer) 109. The traffic policing module is responsible for traffic policing calculations. Normal processing is performed for normal packets, and discarded packets are sent directly to management module 113 without operation. When the traffic policing module 109 is configured to drop mode, the traffic policing module 109 sends the result of the drop/pass along with the packet identifier to the management module 113 after the calculation. When the traffic policing is configured as a shading mode, the packet identifier is calculated based on the "color blind/color sensitive" configuration of the packet. For detailed algorithms, please refer to RFC2697 (A Single Rate Three Color Marker) or RFC2698 (A Two Rate Three Color Marker). The coloring information of the packet identifier is falsified by the result, and the coloring result is sent to the memory unit 114.

Congestion avoidance module 110. The role is to achieve tail drop, WRED or ENC congestion avoidance, Its main functions are: According to the message queue status information, it is determined whether the data packet is discarded. The ECN calculation is performed in ECN mode and the results are given. The WRED algorithm is calculated in WRED mode, and the discard judgment is given. In the tail drop mode, the packet discarding judgment is performed according to the queue depth. The judgment result and the packet identifier information are fed back to the congestion avoidance policy library 103 after the calculation is completed.

Queue scheduling module 111. The function is time slot effective signal generation, back pressure signal generation, packet identifier queue management, scheduling, shaping, back pressure control, providing an interface with the configuration table item management module, and performing parameter configuration.

Traffic interface module 112. The main sub-modules inside the module are as follows: 1. The packet data reading module is mainly responsible for sending the packet identifier readout in the output FIFO to the memory unit 114, and receiving the corresponding packet data to the traffic outbound interface. The main functions are: Collecting the interface FIFO empty condition, and feeding back to the memory unit 114. The packet identifier is sent according to the FIFO condition of the memory unit 114. The read data received from the memory unit 114 is sent to the outgoing interface FIFO based on the received packet port information. Second, the outbound interface FIFO, the role is: The data cache of the sending interface.

2 is a flow chart of the congestion avoidance system of the present invention. The data processing method is as follows: Step A. The packet processor receives the data packet from the upstream device; and enters the traffic interface module 101. Step B. The message classification module 102, using the quintuple of the message, the QOS related information constitutes a key value, the query classification policy library 103, and the discarding priority library 104;

Step C. The message classification module 102 obtains a return value: a classification identifier (or a flow identifier), and discards the priority;

Step D. The message classification module 102 encapsulates the obtained classification identifier, the flow identifier, and the discarding priority into a packet. The TM description information header is placed before the packet;

Step E. The congestion avoidance module 105 forms a query key value according to the classified identifier or the flow identifier sent by the received message classification module 102, and queries the text discard/forward label in the congestion avoidance policy library 106;

Step F. The packet processor congestion avoidance module 105 selects a corresponding forwarding and discarding action according to the packet drop/forward (or drop probability) flag returned by the congestion policy library 106. If it is discarded, discard the current packet and go to step H.

Step G. If the packet is forwarded, the packet processor passes the packet to be forwarded through the traffic interface module 107. Sent to the downstream traffic management device. Go to step I

Step H. The message processing flow ends.

Step I. The traffic interface module 108 of the traffic manager receives the data packet from the upstream packet processing device. Step J: The traffic interface module 108 in the traffic manager obtains the TM description information header before the packet, and obtains the classification identifier or the flow identifier. a specific queue, and performing congestion detection according to a threshold configured by the user; Step K. The traffic manager congestion avoidance module 110 generates a congestion avoidance policy according to the congestion detection result and the drop priority, and the congestion avoidance policy includes a drop probability function, and The packet processing policy in the congestion policy library 106 in the packet processor is updated, and the packet processing policy includes: a packet discarding/forwarding flag;

Step L. The traffic manager performs other actions such as traffic policing, traffic shaping, congestion management, and queue scheduling.

Step M. The traffic manager sends the message to the downstream device through the traffic interface module 112. Those skilled in the art will appreciate that embodiments of the present invention can be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or a combination of software and hardware. Moreover, the invention can be embodied in the form of one or more computer program products embodied on a computer-usable storage medium (including but not limited to disk storage and optical storage, etc.) in which computer usable program code is embodied.

The spirit and scope of the invention. Thus, it is intended that the present invention cover the modifications and variations of the inventions

Industrial Applicability Compared with the prior art, the traffic manager congestion avoidance module generates a congestion avoidance policy according to the congestion detection result and the discarding priority, and the congestion avoidance policy includes the discard probability function, and updates the packet processor. The packet processing policy in the congestion avoidance policy database includes: packet discarding/forwarding flag, and the packet processor and the traffic manager cooperate to complete the congestion avoidance of the packet.

Compared with the existing congestion avoidance mode, when the congestion is discarded, the message is generally not in the flow. Discarding is performed in the volume manager (unless the special case of the TM message queue is full). In the upstream packet processor, the discarding action is performed, and the discarded packet does not need to be forwarded to the TM for processing, which saves the TM resource. Compared with the second method, the invention reduces the memory overhead and the access peripheral overhead caused by the packet processor maintaining a large number of queues, and improves the packet processing performance and efficiency.

Claims

Claim

1. A message congestion avoidance system, the system comprising:

a packet processor, configured to: connect with the traffic manager, complete packet classification, parsing, modification, and encapsulation; perform congestion processing according to the received congestion policy sent by the traffic manager; traffic manager, setting And completing: performing congestion detection of the packet, generating the congestion policy, and sending the congestion policy to the packet processor.

2. The system of claim 1, wherein the packet processor comprises: a message classification module, a classification policy library, and a drop priority library, wherein the text classification module is configured to: query the classification policy library, The discarding priority library obtains a classification identifier and a discarding priority.

The system of claim 2, wherein the packet processor further comprises: a congestion avoidance module and a congestion policy library, wherein the congestion avoidance module is configured to: classify according to the received message classification module Identifying, discarding the priority, querying the congestion policy database, and obtaining the congestion policy.

4. The system of claim 1 or 3, wherein the congestion policy library in the packet processor is configured to: receive the congestion policy fed back by the traffic manager.

5. The system of claim 4, wherein the congestion avoidance module in the traffic manager is configured to: feed back the congestion policy to a congestion policy library in the packet processor.

The system of claim 5, wherein the congestion avoidance module in the traffic manager is further configured to: generate a congestion avoidance policy according to a congestion detection result of the message queue.

7. A method for congestion avoidance, the method comprising:

The traffic manager generates a congestion avoidance policy according to the congestion detection result of the packet queue and the drop priority, and feeds back to the packet processor.

8. The method of claim 7, wherein

The packet processor includes: a message classification module, a classification policy library, and a drop priority library, wherein the message classification module queries the classification policy library, the discarding priority database, and obtains a classification. Identify, discard the priority.

9. The method of claim 7, wherein

The packet processor further includes: a congestion avoidance module and a congestion policy library,

10. The method according to claim 7 or 9, wherein

The congestion policy library in the packet processor receives the congestion policy fed back by the traffic manager; wherein a congestion avoidance module in the traffic manager feeds back the congestion policy to a congestion policy in the packet processor Library.

11. The method of claim 10, wherein

The congestion avoidance module in the traffic manager generates a congestion avoidance policy according to the congestion detection result of the message queue.