KR101037929B1 - Method and Apparatus for packet scheduling - Google Patents

Abstract

A packet scheduling method is disclosed. The present invention is directed to classifying streams entering the scheduler based on speed and / or packet length; For the classified packet, store in the first stream queue if the packet of the stream is the first packet; Store in a second stream queue if it is a subsequent packet; Counting a virtual start service time of a packet stored in a first stream queue according to a weighted fairness queuing method; The virtual start service time of the packet stored in the second stream queue is counted as the virtual service end time of the previous packet. The present invention is a simple and efficient packet scheduling, hardware implementation is easy. It also guarantees the performance of the WF2Q + algorithm.

Description

Packet scheduling method and apparatus {Method and Apparatus for packet scheduling}

1 is a schematic diagram of a scheduling method according to the present invention;

2 is a diagram illustrating an internal configuration of first and second stream queues according to an embodiment of the present invention.

3 is a time flow diagram illustrating a scheduling method according to an embodiment of the present invention.

4 is a time flow diagram illustrating a process of updating a virtual service time of a system between transmission of a previous packet and transmission of a next packet.

5 is a diagram illustrating the structure of first and second stream queues according to another embodiment of the present invention.

6 illustrates a network topology used for simulating the present invention.

7 is a diagram illustrating bandwidth characteristics of a data stream according to one embodiment of FIG. 2.

8 illustrates bandwidth jitter characteristics of a data stream according to an embodiment of FIG. 2.

9 illustrates a delay characteristic of a data stream according to an embodiment of FIG. 2.

FIG. 10 is a diagram illustrating a delay delay characteristic of a data stream according to an embodiment of FIG. 2.

FIG. 11 is a diagram illustrating bandwidth characteristics of a data stream according to one embodiment of FIG. 5. FIG.

12 illustrates bandwidth jitter characteristics of a data stream according to an embodiment of FIG. 5.

13 illustrates a delay characteristic of a data stream according to an embodiment of FIG. 5.

14 illustrates a delay delay characteristic of a data stream according to an embodiment of FIG. 5.

The present invention relates to a network communication system and its application field. The present invention relates in particular to a packet scheduling method applicable to a packet scheduling method of a router.

With the development of the Internet, the multimedia industry is increasingly used. Because the multimedia industry, such as voice and video, relies heavily on bandwidth and delay, routers require increasingly efficient and high-speed packet scheduling methods, which provide reliable quality of service (Qos) between network end devices. Should be provided.                         

Recently, scheduling algorithms based on GPS models such as WFQ, WF ² Q, and WF ² Q + have been widely developed to perform this quality of service guarantee function. The GPS (General Processor Sharing) model is an ideal stream model. This is based on the following assumptions: (1) The length of the packet can be split infinitely (2) All streams can simultaneously accommodate the service. In a real system, the minimum unit serviced by the scheduler server is a packet, and since the scheduler server can service only one stream at the same time, the GPS model is practically impossible to implement. Jon CR Bennett and Hui Zhang developed WF ² Q scheduling algorithms to simulate GPS models in real systems (J. Bennett and H. Zhang, Hierarchical packet fair queuing algorithms, In Proceedings of the ACM-SIGCOMM96 , pages 143-156, Palo Alto, CA, August 1996). The basic idea of this technique is to maintain the start service time and end service time for each packet in the stream. Before the scheduler sends a packet, the scheduler must perform a quality check on the packet to be scheduled. Only packets whose start service time is less than the system virtual time can pass the check. Only those packets that have passed the minimum end service time will be transmitted. This policy is called the Smallest Elgible Virtual Finish time First (SEFF) selection policy.

Since the WF ² Q + method has good fairness and delay characteristics and is not complicated, this algorithm has received widespread attention in the industry. However, the WF2Q + algorithm has the following problems in practical use: First, the complexity of the algorithm increases as the streams to be scheduled increase. Especially for high speed core routers, if the amount of data streams is large, the application of this algorithm will put a heavy load on the system. Second, hardware implementation is not easy.

The present invention seeks to overcome the above problems. The present invention is simple and efficient, and the hardware can be easily implemented. In addition, the present invention guarantees the performance of the WF2Q + algorithm for the service quality assurance function.

In order to achieve the above object, according to an embodiment of the present invention, a packet scheduling method divides and stores a scheduled packet into a first stream queue and a second stream queue, and performs scheduling according to an SEFF policy for each packet. . This process is as follows.

(1) Initialize the scheduling node and set the initial value of the virtual time of the system.

(2) When a predetermined packet reaches the scheduling node, it is checked whether the packet is the first packet of the data stream. If it is the first packet, the packet is stored at the end of the first stream queue Q1 according to the data rate and / or length of the corresponding data stream, and the virtual initiation service time is counted according to equation (2). This time is the virtual initiation service time of the data stream. If it is not the first packet, store the packet directly at the end of the data stream.

(3) In scheduling, the scheduler scans the virtual initiation service time of the head packet of the first data stream in all queues. Thereafter, after detecting a legal packet having a virtual start service time smaller than the system virtual time, the virtual end service time of the legal packet is counted according to Equation (4). The packet is then sent with the minimum virtual termination service time.

(4) The procedure for transmitting the selected head packet is as follows: First. Extract packets from data stream F and send. The data stream F will then be stored at the end of the backlog stream Q2 (Rx, Ly) according to the data rate and length of the new head packet. Then, according to equation (3), the virtual start service time of the data stream F is updated, and this time is the virtual start service time of the new head packet of the data stream F. Thereafter, the virtual time of the system is updated according to the equation (1).

(5) Repeat steps (2) to (4) until the entire operation is completed.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

First, equations and symbols to be used in the following description are defined using Table 1 and Equations 1 to 4.

[Equation 1]

[Equation 2]

&Quot; (3) "

&Quot; (4) "

TABLE 1

V (t) Virtual time function of the system S _i ^k Virtual start time of packet k of data stream i F _i ^k Virtual end time of packet k of data stream i τ System virtual time update time-interval B (t) Set of all streams backlogged to the system at time t Hi (t) The serial number of the head packet of data stream i at time t Qi The amount of scheduled packets in data stream i a _i ^k  Arrival time of packet k in data stream i L _i ^k  Length of packet k in data stream i Ri (t)  Data rate of data stream i at time t

The detailed description of the invention is now described with reference to the associated drawings.

1 is a structural diagram showing a scheduling method according to an embodiment of the present invention.

The scheduler 100 according to the present invention includes a classifier 130, a first stream queue 110, a second stream queue 120, and a SEFF selector 140.

In the scheduler of the present invention, the virtual time function V (t) is defined as in Equation 1 below.                     

[Equation 1]

Where V (t) is the virtual time function of the scheduler 100, τ is the virtual time update interval of the system, B (t) is the set of all streams backlogged to the scheduler 100 at time t, and Hi (t) is The serial number of the head packet of data stream i, _Si ^k is the virtual start time of the k th packet.

Classifier 130 classifies the various data streams according to the data rates. The quantification grades M are assigned in sequence, such as R1, R2, ... Rm. Similarly, the length of the data streams is classified sequentially using L1, L2, ... Ln using the classification class N. For different data streams, the ratings of the length classifications may be the same or may be different. The combination of R and L results in several different data queues, each of which is represented by Q (Rm, Ln). As a result, M * N data queues are obtained, and packets 132 entering the scheduler 100 are classified into different streams according to the speed and packet length of the stream to which the packet belongs.

For a given data rate (R = Rm) and head packet length (L, Ln-1 < L < = Ln), each data stream is stored in the corresponding queue Q (Rm, Ln). This head packet is stored in the stream queues 110 and 120 at the first position at present. In classifier 130, the data length for the classification is selected to be the closest of the lengths greater than its length. Lengths greater than the highest grade are classified as the highest grade. The data streams in the data queue are represented by F1, F2, ..., F _tail , and the packets in the I-th data stream are represented by Pi1, Pi2, ... Pi _{, tail} .

The 'first packet' of a data stream corresponds to a packet that is processed at the beginning of the system or when it is restarted after the system stops at a predetermined time. The method of proving a new packet considers the packet as the first packet when no other packet is waiting for scheduling in the packet's data stream. If there is a packet waiting for scheduling in the data stream of the packet, such a packet is called a 'following packet' of the data stream. Since the count of the virtual start service time of the first packet and the count of subsequent packets of the data stream do not match, the first packet of the data stream must be processed individually. Therefore, as shown in FIG. 1, the data queue is divided into two parts, the first stream queue 110 and the second stream queue 120, and stored. The first stream queue 110 is responsible for the processing of the first packet of the data stream, and the second stream queue is responsible for the processing of subsequent packets of the data stream. Thus, the first and second stream queues 110 and 120 require a total of 2 * M * N queues. The scheduler 140 selects packets at a predetermined time from the first and second stream queues according to the SEFF policy and transmits them to the next node.

2 is a diagram illustrating an internal configuration of first and second stream queues.

The classifier 130 determines which streams of the first stream queue 110 and the second stream queue 120 store packets according to the speed and the packet length of the stream 132 that enters the scheduler 100. At this time, if the packet is the first packet 112, it is stored in the predetermined queue Q1 (R1, L1) in the first stream queue 110, and if the packet is a subsequent packet 122, the second stream queue. The data is stored in the predetermined queue Q2 (R1, L1) in 120. Packets 112 and 122 stored in queue Q1 (R1, L1) and queue Q2 (R1, L1) have the same packet length and stream rate.

3 is a time flow diagram illustrating a scheduling method according to an embodiment of the present invention.

In step 310, the scheduler 100 is initialized and the initial value of the virtual time of the scheduler 100 is set to 0, for example.

In step 320, the classifier 130 checks whether a packet is the first packet of the data stream when the predetermined packet reaches the scheduler 100.

In step 330, if it is the first packet, the packet is stored in the first stream queue Q1 (Rm, Ln) according to the data rate and / or length of the packet.

In step 340, the virtual start service time is then counted according to Equation 2 below. This time is the virtual initiation service time of the data stream.

[Equation 2]

If it is determined in step 320 that the incoming packet is not the first packet of the stream, then in step 332, the packet is stored directly in the second stream queue 120, and then in step 334, Accordingly, the virtual start service time is counted.                     

&Quot; (3) &quot;

In step 350, the SEFF selector 140 scans the virtual start service time for the head packet of the first stream in all queues in the first and second stream queues. In this case, all queues mean M * N first stream queues Q1 (R, L) and M * N second stream queues Q2 (R, L).

Then, in step 360, the SEFF selector 140 detects a packet, that is, a legal packet, whose virtual start service time is smaller than the virtual time of the system.

In step 370, the virtual termination service time of the legal packet is counted according to Equation 4 below.

&Quot; (4) &quot;

Thereafter, in step 380, the SEFF selector 140 transmits the extracted legal packet to the next node at the minimum virtual termination service time. All of the above steps are repeated until all packets have been sent.

4 is a time flow diagram illustrating a process of updating a virtual service time of a system between transmission of a previous packet and transmission of a next packet.

In the case of selecting and transmitting the head packet of the first stream of the first stream queue Q1 (Rm.Ln) or the second stream queue Q2 (Rm.Ln), the following procedures are commonly used in both cases. .

First, in step 410, the SEFF selector 140 extracts the packet from the data stream F and transmits it.

Then, in step 420, a new head packet is generated in a particular first stream queue Q1 (Rx, Lx) or second stream queue Q2 (Rx, Ly) to which the head packet is selected and transmitted. The next packet for the data stream F is also stored in the second stream queue Q2 (Rx.Ly) according to the data rate and length of the new head packet.

Then, in step 430, the virtual start service time of the data stream F is updated according to equation (3). This time is also the virtual initiation service time for the new head packet of data stream F.

Thereafter, in steps 440 and 450, the system virtual time is updated according to equation (1).

Then, in step 460, extraction and transmission for the next packet is performed, after which the service time update by the above steps is performed again.

In one embodiment of the invention, the scheduling node used in the method of the invention is a router.

5 is a view showing the structure of the first and second stream queues according to another embodiment of the present invention.

In FIG. 5, the structure of the second stream queue 520 is the same as in FIG. 2, but the first stream queue 510 is different from FIG. 2. This is to save the queue amount and hardware resources by simplifying the structure of the first stream queue. Referring to FIG. 5, the first stream queue 510 is classified according to either length or speed, which is represented by Q1 (Ln) or Q1 (Rm). Therefore, the queue in the first stream queue 510 needs M * N + N or M * N + M queues.

Hereinafter, FIG. 5 illustrates a scheduling method using a queue using classification by length.

(1) The schedule node is initialized and the initial value of the virtual time of the system is set, for example, 0.

(2) When a predetermined packet arrives at the schedule node, it is checked whether the packet is the first packet of the data stream. If so, the packet is stored at the end of the first stream queue Q1 (Ln) corresponding to that rate, and the virtual start service time is counted according to equation (2). This time is the virtual initiation service time of the data stream. If it is not the first packet, the packet is stored directly at the end of the data stream.

(3) In scheduling, the scheduler scans the virtual initiation service time for the head packet of the first stream in every queue. Here, all the queues refer to N first stream queues Q1 (L) and M * N second stream queues Q2 (R, L). Then, a legal packet having a virtual initiation service time smaller than the system virtual time is extracted, and the virtual termination service time of the legal packet is counted according to Equation (4). The packet is then sent at the minimum virtual termination service time.

(4) In the case of selecting and transmitting the head packet of the first stream of the first stream queue Q1 (Ln) or the second stream queue Q2 (Rm.Ln), the following procedure is used in both of the above cases. First, a packet is extracted from the data stream F and transmitted. The data stream F then stores the packet at the end of the second stream queue Q2 (Rx.Ly) according to the data rate and length of the new head packet. Then, according to equation (3), the virtual start service time of data stream F is updated, which is also the virtual start service time of the new head packet of data stream F. Thereafter, the system virtual time is updated according to Equation 1.

(5) Steps (2) to (4) are repeated until the entire scheduling is completed.

6 is a diagram illustrating a network topology used for simulating the present invention.

In Figure 6, the bandwidth of each input chain is 10M. All data streams are scheduled by the scheduling node and output to the output chain. The speed and length of the schedule node of FIG. 6 were used in the simplified method of FIG. 5 and classified into five classes.

The five speed classes of the data stream used in this simulation are as follows:

0.1 Mbps, 0.3 Mbps, 1 Mbps, 2 Mbps, 5 Mbps

The five length classes of the data stream used in this simulation are as follows.

200, 400, 800, 1000, 1600bytes

A data stream transmission method applicable to the present invention is as follows. In this simulation, an ON / OFF method is used.                     

-Constant Bot Rate (CBR) method: Uses a data stream with a constant rate.

-On / Off method: Use the data stream transmitted intermittently.

Same packet length: All data packets in the data stream use the same data stream.

Average Distribution scheme: The length of packets in the data stream is evenly distributed within a predetermined range.

Normal Distribution: The length of a packet in a data stream is normally distributed in a predetermined range based on a predetermined center value.

7 to 14 are diagrams showing the results of simulation by the scheduling of the present invention. To demonstrate the performance of the packet scheduling method according to the present invention, it was simulated using a network simulator and the following performance indices were tested.

Bandwidth: The actual bandwidth (Mbps) received by each data stream.

* Bandwidth jitter: the average actual difference in Mbps of the bandwidth received by each data stream in an adjacent period

Delay: the difference between the departure and arrival times of packets in each data stream (ms)

Delay jitter: Average delay difference (ms) between the previous and subsequent packets of each data stream.

The results refer to some of the detailed performance test methods. Overall, this method is simple, efficient, and easy to implement hardware. The present invention also ensures the performance of the WF2Q + algorithm.

The parameters used in this simulation are shown in Table 2.

TABLE 2

Data stream Chain bandwidth Bandwidth performed Transmission method and speed Packet length One 10 Mbps 5.0 Mbps 50% On / Off 5.0 Mbps Normal distribution 2 10 Mbps 2.0 Mbps 20% On / Off 2.0 Mbps Normal distribution 3 10 Mbps 1.0 Mbps 10% On / Off 1.0 Mbps Normal distribution 4 10 Mbps 1.0 Mbps 10% On / Off 1.0 Mbps Normal distribution 5 10 Mbps 0.3 Mbps 3% On / Off 0.3 Mbps Normal distribution 6 10 Mbps 0.3 Mbps 3% On / Off 0.3 Mbps Normal distribution 7 10 Mbps 0.1 Mbps One% On / Off 0.1 Mbps Normal distribution 8 10 Mbps 0.1 Mbps One% On / Off 0.1 Mbps Normal distribution 9 10 Mbps 0.1 Mbps One% On / Off 0.1 Mbps Normal distribution 10 10 Mbps 0.1 Mbps One% On / Off 0.1 Mbps Normal distribution Scheduling output 10 Mbps

On / Off models were used for each of the data streams of the above configuration. Packets in the data stream have a normal distribution (average is 1000 bits, deviation is 400 bits). In this test, the Performance Index of bandwidth, bandwidth jitter, delay, delay jitter was measured for the two methods shown in FIGS. 2 and 5.

7 to 10 illustrate bandwidth, bandwidth jitter, delay and delay jitter according to the method of FIG. 2, and FIGS. 11 to 14 illustrate bandwidth, bandwidth jitter, delay and delay jitter according to the method of FIG. 5.

According to the simulation results of FIGS. 7 to 14, it can be seen that the two scheduling methods according to the present invention guarantee all four performance indices. Therefore, the quality of service of the customer can be guaranteed.                     

Based on the simulation of the network simulator (NS), Xilinx Field Programmable Gate Array (FPGA) was used as the scheduling chip. The chip supports data streams up to 128K and five speed and length classes. The chip is also available in a variety of speed and length grades. Based on actual processing tests, the chip was able to guarantee all the required bandwidth, delay and fairness of each data stream.

On the other hand, the scheduling method according to the present invention can be created by a computer program. The codes and code segments that make up the program can be easily deduced by a computer programmer in the field. The program is also stored in a computer readable media, which is read and executed by a computer to implement scheduling of packets. The information storage medium includes a magnetic recording medium, an optical recording medium, and a carrier wave medium.

So far I looked at the center of the preferred embodiment for the present invention. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

 As described above, according to the present invention, even if the number of streams increases, packet scheduling using a simple hardware structure is possible.

Claims (20)

In the packet scheduling method, a) classifying streams entering the scheduler based on rate, packet length, or rate and packet length; b) for the classified packet, storing in the first stream queue if the packet of the stream is the first packet and storing in the second stream queue if it is a subsequent packet; c) counting a virtual start service time of the packet stored in the first stream queue according to a weighted fairness queuing method; And d) counting a virtual start service time of the packet stored in the second stream queue as a virtual service end time of a previous packet; Method comprising a. The method of claim 1, characterized in that step c) is carried out by ^{WF 2 Q (Worst-case Fair} Weighted Fair Queueing) scheduling algorithm or ^{WF 2 Q + (Worst-case} Fair Weighted Fair Queueing +) Scheduling Algorithm . The method of claim 2, wherein c) is represented by the following equation,

S _i ^k is the virtual initiation service time of the k th packet of the I th stream, V (t) is the virtual time function of the system, a _i ^k is the arrival time of the k th packet of the I th stream, F _i ^k-1 is the virtual termination service time of the k-1th packet of the Ith stream, and Qi is the amount of previous packets contained in the corresponding queue of the Ith stream. 2. The method of claim 1, wherein step d) is performed by a Smallest Eligible virtual Finish time First (SEFF). The method of claim 1, further comprising: e) scanning a virtual start service time of a packet stored in the first stream queue and the second stream queue to detect a legal packet whose virtual start service time is less than a virtual service time of a system. It further comprises a method. The method of claim 5, wherein step e) e1) the virtual termination service time of the legal packet is represented by the following equation,

Counting according to the equation, wherein F _i ^k is the virtual end service time of the k th packet of the I th stream, S _i ^k is the virtual start service time of the k th packet of the I th stream, and L _i ^k is the I th The length of the k th packet of the stream, Ri (t), represents the speed of the I th stream. 6. The method of claim 5, comprising: f) sending the detected legal packet to a next node. The method of claim 1, wherein the first stream queue of step b) is And classify according to the speed of the stream. The method of claim 1, wherein the first stream queue of step b) is And classify according to the packet length of the stream. The method of claim 1, wherein the first stream queue of step b) is And classify according to the speed and packet length of the stream. A packet scheduling apparatus using a weighted fair queue method, a) a classifier that classifies streams entering the scheduler based on rate, packet length, or rate and packet length; b) a first stream queue storing a first packet of the classified packets; c) a second stream queue for storing subsequent packets of the classified packets; And d) a SEFF selector for detecting a legal packet according to a SEFF policy for all packets stored in the first stream queue and the second stream queue. The method of claim 11, wherein the SEFF selector, A virtual start service time of the packet stored in the first stream queue is counted according to a weighted fairness queuing method, And count the virtual initiation service time of the packet stored in the second stream queue as the virtual service end time of the previous packet. The virtual time function of claim 12, wherein:

Where V (t) is the virtual time function of the scheduler 100, τ is the virtual time update interval of the system, B (t) is the set of all streams backlogged to the scheduler 100 at time t, Hi ( t) is the serial number of the head packet of data stream i, S _i ^k is the virtual start time of the k th packet. The method of claim 12, wherein the SEFF selector is,

Counts the virtual initiation service time of the first stream queue, where S _i ^k is the virtual initiation service time of the k th packet of the I th stream, V (t) is the system's virtual time function, and a _i ^k is I The arrival time of the k-th packet of the first stream, F _i ^k-1 is the virtual termination service time of the k-1th packet of the I-th stream, and Qi represents the amount of previous packets in the corresponding queue of the I-th stream. Device. The method of claim 12, wherein the SEFF selector is,

Counts the virtual initiation service time of the second stream queue, where S _i ^k is the virtual initiation service time of the k th packet of the I th stream, and F _i ^k-1 is the k-1 th packet of the I th stream Wherein the virtual termination service time of &lt; RTI ID = 0.0 &gt; Qi &lt; / RTI &gt; represents the amount of previous packets contained in the corresponding queue of the I-th stream. The apparatus of claim 12, wherein the SEFF selector is configured to scan a virtual start service time of packets stored in the first stream queue and the second stream queue to detect a legal packet having a virtual start service time smaller than a virtual service time of a system. Device characterized in that. The method of claim 16, wherein the SEFF selector is,

Count the virtual termination service time of the legal packet according to F _i ^k , the virtual termination service time of the k-th packet of the I-th stream, S _i ^k , the virtual initiation service time of the k-th packet of the I-th stream, and L _i ^k is the length of the k th packet of the I th stream, and Ri (t) is the speed of the I th stream. The method of claim 11, wherein the first stream queue, And classify according to the speed of the stream. The method of claim 11, wherein the first stream queue, And classify according to the packet length of the stream. The method of claim 11, wherein the first stream queue, And classify according to the speed and packet length of the stream.

Images