WO2006069528A1

WO2006069528A1 - A packet scheduling method in the packet service

Info

Publication number: WO2006069528A1
Application number: PCT/CN2005/002312
Authority: WO
Inventors: Wumao Chen; Xueyi Zhao
Original assignee: Huawei Technologies Co., Ltd.
Priority date: 2004-12-29
Filing date: 2005-12-26
Publication date: 2006-07-06
Also published as: CN100370787C; CN1798090A

Abstract

A packet scheduling method in the packet service includes: dividing users into the high priority user and the low priority user; determining the minimum bandwidth of the high priority user, and setting its flow rate measure; measuring the flow of the packet of the high priority user, then the packet being not out of the flow rate measure is the high priority packet and the packet being out of the flow rate measure is the general packet; regarding the packet of the general user as general packet; putting the high priority packet in the transmitting queue for transmission; for general packet, detecting the congestion condition of the output: discarding the excess general packet if the output was congested, and transmitting the general packet if the output was not congested. The present invention ensures the minimum bandwidth of the high priority user flow, whilst, all the users share the remainder bandwidth to completely provide the required bandwidth for UBR+ service.

Description

Packet scheduling method in packet service

TECHNICAL FIELD The present invention relates to a scheduling method, and more particularly to a scheduling method applicable to UBR+ (enhanced UBR, in which UBR is an Unspecified Bit Rate, i.e., "unspecified bit rate" abbreviation) in a packet service system.

Background technique

In a packet service system, quality of service (QoS) parameters include packet delay, delay jitter, loss rate, and throughput. Etc. Different services require the network to provide different QoS parameters, and different QoS parameters need to be guaranteed by corresponding scheduling methods. The so-called scheduling, also known as flow control, is to control the order of the connected services to achieve the following objectives: 1) to ensure that each connection enjoys the bandwidth of its reservation; 2) the remaining bandwidth in the system In the case of the remaining bandwidth, the remaining bandwidth is allocated according to the QoS requirements of each connection; 3) The delay requirements of each connection are guaranteed.

The service goals of the existing UBR+ service QoS requirements are as follows: 1) Ensure the connection bandwidth with minimum bandwidth requirement, and the rest does not require the connection to do its best; 2) For the remaining bandwidth of the system, all connections are shared, including Connections with minimum bandwidth requirements; 3) Do their best in terms of latency.

Currently, there are two main scheduling algorithms commonly used for this service: the first is the rim and fair scheduling algorithm, and the second is the scheduling algorithm with Strict-Priority Queue (abbreviated as PQ). . Before describing the two methods, the following definitions are made: For a connection with minimum bandwidth guarantee, it is called high priority connection, for a connection without bandwidth requirement, it is called normal connection; The corresponding user is called high priority. Level users as well as regular users.

Rim and fair scheduling are the most commonly used scheduling algorithms, including round robin algorithm (RR), eighted Round Robin (abbreviated as WRR), and DRR (abbreviation of Deficit Round-Robin). Although these algorithms can well guarantee the fairness of bandwidth allocation among users, or can ensure that each user shares the export bandwidth proportionally; In the following deficiencies: It is difficult to guarantee the minimum bandwidth of one or some high-priority users while the remaining bandwidth is shared by all users. In order to ensure that all users share the remaining bandwidth while ensuring high-priority user bandwidth, it is necessary to dynamically modify the scheduling weight of each user according to the number of connections of the user. For example, suppose the egress bandwidth is N. When there is a high-priority user, in order to achieve the UBR+ bandwidth requirement, the minimum bandwidth required for the user is guaranteed to be M, and the user and other L ordinary users are simultaneously assumed (assuming weights) To 1) be able to share the remaining bandwidth (NM), then the user's weight needs to be set to M*(L+l) / (N-M). However, in actual applications, the number of high-priority users and the number of ordinary users may be randomly released, so it is cumbersome to calculate and set the weight of each user. More importantly, due to the fact that the number of user connections is variable at any time, this will make it difficult or even impossible to accurately set the weight values of each user.

Another common scheduling method is to implement bandwidth guarantee and sharing of UBR+ services by using traffic policing and PQ scheduling. The traffic policing (Commi t Acces s Rate, abbreviated as CAR) sets an allowable traversal rate. Traffic within this rate range can pass, and traffic that exceeds this rate is discarded. Moreover, in order to ensure that each high-priority user can obtain the minimum required bandwidth, the sum of the minimum bandwidth required by all high-priority users when using traffic policing cannot exceed the total outgoing bandwidth. At the same time, PQ scheduling is used in this scheduling method to ensure that all high-priority user traffic that passes traffic policing always takes precedence over ordinary user traffic. The working principle is shown in Figure 1. In Figure 1, the packet 1 of the high-priority user passes the traffic policing 2, allowing the traffic in the rate range to pass through and entering the high-priority traffic queue 3; Packet 4 enters the normal traffic queue 5; then, through PQ scheduling 6, it ensures that high-priority user traffic takes precedence over ordinary user traffic. In Figure 1, although the minimum bandwidth of high-priority user traffic is guaranteed, only ordinary users share the remaining bandwidth, and those that exceed the minimum bandwidth of those high-priority user traffic are directly discarded and are not shared with ordinary user traffic. Remaining bandwidth. Therefore, the above solution does not meet the bandwidth requirements of the UBR+ service.

In summary, regardless of the rim and fair scheduling algorithm, or the traffic policing plus absolute priority scheduling algorithm, only the minimum bandwidth requirement of high-priority users can be guaranteed, and it is difficult or impossible to achieve high-priority users and ordinary users. Sharing the remaining bandwidth, unable to meet the bandwidth requirements of the UBR+ service.

Summary of the invention The technical problem to be solved by the present invention is to provide a scheduling method for a single cartridge, which can solve the minimum bandwidth of one or some high-priority users, and can also achieve the remaining bandwidth by including higher priority users. The problem shared by all users can fully meet the bandwidth requirements of the UBR+ service.

The technical solution adopted by the present invention to solve the technical problem thereof is as follows: A packet scheduling method in a packet service is provided, which includes the following steps:

S1: User classification, dividing all users into high priority users and ordinary users;

S2: determining a traffic metric rate of each high priority user;

S3: Perform traffic metrics on the data packets of the high-priority user, and use the data packet that does not exceed the traffic metric rate as the high-priority data packet, and use the data packet that exceeds the traffic metric rate and the data packet of the ordinary user as the common data pack;

S4: For the high priority data packet, send it to the sending queue of the egress; for normal data packets, check whether the egress is congested, and if the egress is congested, discard the excess normal data packet, otherwise the ordinary data packet is placed. Send out to the send queue of the exit.

The sending queue includes a high-priority traffic queue and a normal traffic queue. The high-priority data packet is sent to the high-priority traffic queue, and all the ordinary data packets are sent to the normal traffic queue for transmission.

The absolute priority scheduling is also adopted in the step S4, so that the high priority traffic queue always takes precedence over the normal traffic queue.

Alternatively, the sending queue is implemented by using a FIFO queue, and the FIFO queue is provided with a high threshold and a low threshold; the high priority data packet is judged to enter the queue according to the high threshold threshold, and is discarded when the queue cannot be entered; The normal data packet is judged to enter the queue according to the low threshold threshold, and is discarded when the queue cannot be entered. The difference between the high threshold and the low threshold is not less than the sum of the bursts of all the high priority users.

Alternatively, the high-priority traffic queue and the common traffic queue are respectively set with respective thresholds, and the high-priority data packets exceeding the high-priority queue threshold are marked as low-priority data packet processing, and the low-priority queue threshold is exceeded. The low priority packets are discarded.

The packet scheduling method in the packet service of the present invention performs traffic metrics on data packets of high-priority users, and divides data packets that do not exceed the traffic metric rate into high-priority data. Packet, which divides the data packet exceeding the traffic metric rate into ordinary data packets; and divides the ordinary user data packet into ordinary data packets; sends the high priority data packets into the queue, and controls the entry of the ordinary data packets when the egress is blocked. The team ensures the minimum bandwidth guarantee for high-priority user traffic, and all users share the remaining bandwidth, which can fully meet the bandwidth requirements of UBR+ services.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a schematic diagram of a working principle of scheduling using traffic policing and PQ.

2 is a flow chart of a packet scheduling method in the packet service of the present invention.

3 is a schematic diagram showing the operation of a packet scheduling method in the packet service of the present invention.

4 is a schematic diagram of a first embodiment of a packet scheduling method in a packet service of the present invention. FIG. 5 is a schematic diagram of a second embodiment of a packet scheduling method in a packet service according to the present invention. 6 is a schematic diagram of a third embodiment of a packet scheduling method in a packet service of the present invention. detailed description

The packet scheduling method in a packet service of the present invention will be further described in detail below with reference to the accompanying drawings.

As shown in FIG. 1 and FIG. 3, the packet scheduling method in the packet service of the present invention is first in steps.

S1 performs user classification, that is, divides all users into two categories: high priority users and ordinary users; then, determining the minimum bandwidth of each high priority user in step S2, thereby setting the traffic measurement rate; then, in step S3, The data packet sent by the user is classified, that is, the traffic metric of the high-priority user is measured according to the traffic metric rate set in step S2, and the data packet of the high-priority user may be checked by using a leaky bucket algorithm or a token bucket algorithm. 10 is the excess traffic, the packet that does not exceed the traffic metric rate is divided into the high priority packet 101, the packet exceeding the traffic metric rate is divided into the ordinary data packet 102; and the ordinary user data packet ^{20 is} divided into ordinary packets. Packet 202. Finally, the data packet is processed at the exit of step S4, the high priority data packet 101 is placed in the sending queue of the egress, and sent out; the normal data packet 102, 202 is checked for the congestion status of the egress, and if the egress is congested, the packet is discarded. The excess normal data packets 102, 202, if the exit is unblocked, put the ordinary data packets 102, 202 into the outgoing transmission queue and send them out. This ensures the minimum bandwidth guarantee for high-priority user traffic, and at the same time enables all users to fully share the remaining bandwidth.

Referring to FIG. 4, in a first embodiment of a packet scheduling method in a packet service of the present invention, First divide all users into high-priority users and ordinary users; then determine the minimum bandwidth of each high-priority user, and then set the traffic metric rate; then perform traffic metrics, packet classification. In this embodiment, the traffic metric only performs the measurement of the rate without discarding, that is, the data packet 10 of the high priority user is classified, and the data packet within the rate requirement range is divided into the high priority data packet 101, and the packet is entered. To the high-priority traffic queue; divide the data packet required by the original over-rate into the normal data packet 102, and use the same enqueue control as the ordinary data packet 20 of other ordinary users to enter the normal traffic queue. And the PQ scheduling module is used for scheduling at the exit to ensure that all high priority packets always take precedence over ordinary packets.

Referring to FIG. 5, in the second embodiment of the packet scheduling method in the packet service of the present invention, all users are first divided into high-priority users and ordinary users; then the minimum bandwidth of each high-priority user is determined, thereby setting Traffic metric rate; then traffic metrics, packet classification, classifying packets 10 of the high-priority user in the rate range into high-priority packets

101. The data packet 10 exceeding the rate range is divided into the ordinary data packet 102, 202 together with the data packet 20 of the ordinary user. The FIFO (First In First Out) queue is then used at the exit for packet processing. The FIFO queue is provided with a high threshold and a low threshold; the high priority data packet 101 is judged to enter the queue according to the high threshold threshold, and is discarded when the queue cannot be entered; the common data packet

102, 202 The threshold value of the low threshold is judged and entered into the queue.

The FIFO queues described here, as well as the high and low thresholds, actually involve another implementation. The embodiment of Figure 4 uses two queues, one with a higher priority and one with a lower priority, while the embodiment of Figure 5 uses only one FIFO queue, but simulates by setting two thresholds in this FIFO queue. 2 logical queues (physically only one queue). The scheduling of high and low priority packets is achieved through these two logical queues. For example, a FIFO queue depth is up to 1000 packets, assuming a low priority threshold of 500, a high priority threshold of 1000, and the number of packets stored in this FIFO queue is less than 500. When high- and low-priority data packets can be enqueued, when the data packets stored in the FIFO are greater than 500 and less than 1000, only high-priority data packets can be enqueued, and low-priority data packets are Discarded; when the packet stored in this FIFO queue reaches 1000 (full), the high-priority packet cannot be enqueued and discarded. In this way, a physical queue is used to schedule two priority packets.

In FIG. 5, since the threshold of the high threshold is greater than the threshold of the low threshold, the high priority data packet is prioritized over the ordinary data packet, and the minimum bandwidth of the high priority user is guaranteed; The data packet of the out-of-rate range and the data packet of the ordinary user form a common data packet, so that the high-priority user and the ordinary user can share the remaining bandwidth. When the congestion occurs at the egress, in order to ensure that the bandwidth requirements of all high-priority users can be guaranteed, the difference between the thresholds of the high-gate P艮 and the low threshold is not less than the sum of the bursts of all the high-priority users, that is, Burst data of all high-priority users can enter the queue, thus ensuring the bandwidth requirements of high-priority users. The reason why the difference between the high and low thresholds is not less than the sum of the bursts of all high-priority users is to ensure that high-priority packets of multiple user bursts cannot be cached (discarded) because the queue capacity is too small. For example, if there are 100 high-priority users, the maximum rate of these users is 10 packets in 1 second, and our FIFO queue export processing capability is 800 packets in 1 second. For a limit case, Low-priority data packets already occupy all the memory cells in the low-limit threshold of the FIFO queue, so as long as there are 200 (1000 - 800) memory cells between the high and low thresholds, it will not cause the user to send data at the same time. Packets cause packet loss.

Referring to FIG. 6, in the third embodiment of the packet scheduling method in the packet service of the present invention, all users are first divided into high-priority users and ordinary users; then the minimum bandwidth of each high-priority user is determined, thereby setting The traffic metric rate is then divided into a high priority packet 101, and the packet 10 exceeding the rate range is divided into the normal packet 102. There are high priority traffic queues and normal traffic queues at the exit. High-priority packets 101 are placed in the high-priority traffic queue, and ordinary users' packets are placed in the normal traffic queue. The high-priority traffic queue and the common traffic queue are respectively set with respective thresholds, and the high-priority data packets exceeding the high-priority queue threshold are not discarded, and they are marked as low-priority data packet processing. At this time, these high-priority data packets become low-priority data packets, and their processing is completely the same as low-priority data packets, and no distinction is made; for low-priority data packets exceeding the low priority queue threshold. Discard processing. The low priority packet at this time may include two parts, one of which is a low priority packet, and the other is a low priority packet that is changed by a high priority packet that cannot enter the high priority queue. The PQ scheduling module is used for scheduling at the exit to ensure that all high priority packets always take precedence over normal packets. This ensures the minimum bandwidth guarantee for high-priority user traffic, and all users share the remaining bandwidth, which can fully meet the bandwidth requirements of UBR+ services.

Claims

Rights request

1. A packet scheduling method in a packet service, comprising the following steps:

S2: determining a traffic metric rate of each high priority user;

It is characterized in that it further comprises the following steps:

S3: Perform traffic measurement on the data packet of the high-priority user, and use the data packet that does not exceed the traffic metric rate as the high-priority data packet, and use the data packet that exceeds the traffic metric rate and the data packet of the ordinary user as the common data pack;

S4: For the high priority data packet, send it to the sending queue of the egress; for the normal data packet, check whether the egress is congested, and if the congestion is over, discard the excess normal data packet, otherwise the ordinary data packet is put Send out to the send queue of the exit.

The packet scheduling method of the packet service according to claim 1, wherein the sending queue comprises a high priority traffic queue and a common traffic queue, and the high priority data packet passes the high priority The traffic queue is sent, and the normal data packet is sent through a common traffic queue.

The packet scheduling method in the packet service according to claim 1, wherein the priority of the high priority traffic queue is prioritized over the priority of the normal traffic queue.

The packet scheduling method in the packet service according to claim 1, wherein the sending queue is a FIFO queue, the FIFO queue is provided with a high threshold and a low threshold; and the high priority data packet is based on The high threshold value is judged to enter the queue, and the discarding process is not performed when the queue is not entered. The common data packet is judged to enter the queue according to the low threshold threshold, and is discarded when the queue cannot be entered.

The packet scheduling method in the packet service according to claim 4, wherein the difference between the thresholds of the high threshold and the low threshold is not less than the sum of the bursts of all the high priority users.

The data packet scheduling method in the packet service according to claim 2 or 3, wherein the high priority traffic queue and the normal traffic queue are respectively set with respective thresholds, which will exceed the high priority queue threshold. High-priority packets are marked for low-priority packet processing, for low-priority congestions that exceed low-priority queue thresholds