JP2008193310A - Buffer management method of router and router using the management method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve throughput performance of a communication network by effectively suppressing generation of RTO by actively suppressing discard of a plurality of packet data in the same window and discard of resent packet data when transferring packet data. <P>SOLUTION: Packet data out of those data made discard candidates by an RED function of the router which are managed every flow to which the packet data belong, whose preferential count decreasing in value when fetched by a buffer of the router does not satisfy a discard condition, and which are resent are not discarded; and packet data out of those data input in succession to the router which have the preferential count satisfy the discard condition and are not resent are discarded. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a router buffer management method and a router using the management method. More specifically, the present invention relates to a technique for controlling discarding of packet data from a router buffer due to communication network congestion.

  In this specification, packet data is used to mean data transferred in a communication network and divided into packets. The flow is used to mean one-to-one communication between the transmission side and the reception side. The window is used to mean an amount for determining the number of packets transmitted to the communication network at a time.

  A router, which is a communication device that constitutes a communication network, has a buffer. When a large amount of packet data exceeding the processing capacity is transmitted at a time, the packet data is temporarily stored in the buffer to reduce traffic. Absorb fluctuations. However, if the amount of traffic and the number of users increase more than expected and the amount of input packet data exceeds the buffer capacity of the router, the packet data overflows from the buffer and is discarded.

  The congestion control of the communication network is performed by a transmission control protocol (hereinafter referred to as TCP) function that is used by most applications as a data transfer protocol. TCP detects the discard of packet data in the buffer of the router at the transmitting end, and performs control to reduce the amount of packet data sent to the communication network by reducing the window size that determines the amount of packets that can be transmitted at once. Reduce congestion. TCP also resends discarded packet data.

  Specifically, when two or more packet data are discarded in the same window of a single flow, TCP waits for a certain period of time to send packet data, then returns the window size to 1 and starts retransmission. Retransmission Time Out (Hereinafter referred to as RTO) is performed. In addition, when packet data is discarded over a plurality of flows, the TCP synchronizes and reduces a plurality of window sizes related to these flows, thereby greatly reducing the transmission capacity of the communication network. The amount of transmission is reduced, and the throughput of the communication network, that is, the transmission amount of communication data decreases.

  Conventionally, an active queue management method has been proposed to solve the problem of data transmission delay due to the occurrence of RTO, that is, the waiting time and the problem that TCP operates synchronously and reduces a plurality of window sizes. Yes. A typical example is Random Early Detection (hereinafter referred to as RED). RED probabilistically discards packet data that is input to the router before the buffer is full in order to prevent the buffer from overflowing and continuously discarding packet data that is input to the router. As a result, the amount of TCP packet data transmitted is controlled, and an excessive congestion state of the communication network is avoided.

  As a conventional packet data transmission control method utilizing the RED function, for example, there is a network system that transmits packet data (Patent Document 1). As shown in FIG. 6, this network system is a variable that is provided at the entrance of a network and that indicates the lifetime of packet data that has entered the network, and decreases as the number of nodes that have passed through the network system increases. When the time-to-live (hereinafter referred to as TTL) field value is set to a predetermined value and the packet data 108 to be transferred is input, packet data having a small TTL field value is A flow discriminator 103 having a TTL discriminator 102 that allocates packet data having a large TTL field value to a queue having a low priority among the queues 104, 105, and 106, and assigns the packet data having a large TTL field value to the queue having a low priority among the queues 104, 105, and 106; Assigned to a high priority queue Composed of at least one node 101 Metropolitan with a scheduler 107 from the obtained packet data performs processing to output to the next node after the TTL field value by subtracting 1 in this order. Then, by considering the TTL field value in discarding packet data by the RED function, the discard rate of packet data that has passed through more nodes is reduced.

JP 2003-348140 A

  However, the network system of Patent Document 1 does not have a mechanism for reducing the discard rate of multiple packet data in the same window or the discard rate of retransmitted packet data. For this reason, it is not possible to effectively suppress the occurrence of RTO due to the discard of multiple packet data within the same window or the discard of retransmitted packet data, and the throughput performance of the communication network can be sufficiently improved. It's hard to say.

  Therefore, the present invention effectively suppresses the occurrence of RTO and improves the throughput performance of the communication network by actively suppressing the discarding of a plurality of packet data within the same window and the discarding of retransmitted packet data. It is an object of the present invention to provide a router buffer management method capable of performing the above and a router using the management method.

  In order to achieve this object, the router buffer management method according to claim 1 is managed for each flow to which packet data belongs among packet data determined as discard candidates by the RED function of the router, and is taken into the buffer of the router. The preferential count for which the value is reduced does not satisfy the discard condition and the retransmitted one is not discarded, and the preferential count satisfies the discard condition and is retransmitted from the packet data input after the router. I try to dispose of things that I don't have.

  According to another aspect of the present invention, there is provided a router for managing whether or not to discard packet data by a RED function, and a discard flag management for managing a discard flag whose value changes based on a determination result of the RED execution unit. And a discard flag determination unit that determines whether or not the packet data is a discard candidate based on the value of the discard flag, and a preferential count management that manages a preferential count that decreases when the packet data is loaded into the buffer for each flow. Means, preferential count determination means for determining whether or not the preferential count satisfies the discard condition, reproduction packet data management means for managing the sequence number of the packet data for each flow, and a sequence number managed for each flow And retransmission packet data determination means for determining whether the packet data has been retransmitted based on I have to so that.

  Therefore, according to the buffer management method of this router and the router using the management method, the same flow can be obtained by preferentially not discarding a certain number of packet data of the same flow based on the preferential count of the flow to which the packet data belongs. The discarding of multiple packet data is suppressed. Further, discarding of retransmitted packet data is suppressed.

  Furthermore, since the router buffer management method of the present invention operates in cooperation with conventional TCP by providing a function to the router in the same manner as RED that has been used in the past, it is compatible with the terminal side of a communication network. It is not necessary to have the function to do.

  According to the buffer management method of the router of the present invention and the router using the management method, the discard of a plurality of packet data of the same flow is suppressed and the discard of the plurality of packet data of the same window is suppressed and retransmitted. Since discarding of packet data is suppressed, the occurrence of RTO can be effectively suppressed, and the throughput performance of the communication network can be improved.

  Furthermore, according to the buffer management method of the router of the present invention, since it is not necessary to provide a function corresponding to the terminal side of the communication network, there is an advantage that it can be easily implemented.

  Hereinafter, the configuration of the present invention will be described in detail based on the best mode shown in the drawings.

  1 to 3 show an example of an embodiment of a router buffer management method of the present invention. In this buffer management method of the router, the preferential count whose value is reduced when it is managed for each flow to which the packet data belongs out of the packet data that is considered as a discard candidate by the router RED function and taken into the router buffer satisfies the discard condition. The packets that have not been retransmitted and the ones that have been retransmitted are not discarded, but the packet data that is input following the router is discarded if the preferential count satisfies the discard condition and has not been retransmitted.

  The router buffer management method is realized as a router using the buffer management method of the present invention. The router of the present embodiment includes a RED execution unit that determines whether or not to discard packet data by the RED function, a discard flag management unit that manages a discard flag whose value changes based on a determination result of the RED execution unit, A discard flag determining means for determining whether or not the packet data is a discard candidate based on the value of the discard flag; a preferential count managing means for managing a preferential count for which the value is reduced for each flow when the packet data is taken into the buffer; Preferential count determination means for determining whether the preferential count satisfies the discard condition, a reproduction packet data management means for managing the sequence number of the packet data for each flow, and a packet based on the sequence number managed for each flow Having retransmission packet data determination means for determining whether or not the data has been retransmitted A.

  In executing the router buffer management method of the present invention, first, packet data is input to a router installed on a communication network (S1).

  The communication network targeted by the present invention is specifically a TCP / IP communication network. Hereinafter, the TCP / IP communication network is simply referred to as a network.

  In the present embodiment, a description will be given of a case where the TCP Reno version implemented in most operating systems is used as the TCP among various versions of TCP.

  TCP itself has a function of suppressing excessive network congestion. Specifically, the TCP Reno version controls the amount of packet data transmitted by changing the amount of packet data that can be transmitted to the network at once by a congestion window (Congestion Windows) (for example, WR Stevens; TCP / IP Illustrated, Volume 1: The Protocols, Addison Wesley Longman, see 1994).

  The TCP on the transmission side increases the window size every time an acknowledgment packet for confirming that data has been transmitted normally is received from the reception side. The increase width starts from one packet and increases exponentially as 2, 4,..., But when the packet size is discarded, if the window size exceeds the threshold automatically set to the previous half, the increase width is 1 packet. It becomes one by one. The former is called a slow start phase, and the latter is called a congestion avoidance phase. The increase in the window size is continued until packet data is discarded due to network congestion.

  There are two types of packet data discarding detection methods and subsequent retransmission operations: RTO and Fast Retransmission. RTO is a method for detecting packet data discarding when a time until an acknowledgment packet arrives due to discarding of a plurality of packet data in the same window exceeds a timer setting time. In this case, TCP waits for a certain period of time to send packet data, then returns the congestion window size to 1 packet, and starts retransmission from slow start.

  On the other hand, Fast Retransmission detects the discard of packet data by discarding single packet data in the same window and receiving an acknowledgment packet with the same number indicating the packet to be transmitted next three times. It is a method to do. In this case, the TCP determines that the congestion state is slight, and immediately starts retransmission by reducing the congestion window size to the previous half.

  In the present embodiment, a router that has at least a buffer, a RED function, and a layer 4 processing capability and is used as a relay device for connecting networks is used.

  Subsequently, the RED executing means of the router determines whether or not to discard the packet data input to the router by the RED function (S2). Note that the arithmetic unit and the control unit of the router that actually perform the calculation and various processes related to the buffer management method of the present invention are, for example, a CPU (Central Processing Unit). In the present specification, it is simply expressed as a router, or as means configured in a calculation unit or a control unit.

  Since the RED function itself is a well-known technique, details are omitted here (for example, S. Floyd and V. Jacobson; Random early detection gateways for congestion avoidance, IEEE / ACM Trans. Netw., Vol. 1, no. 4, pp. 397-413, August 1993). The RED function is mounted on the router.

  A specific algorithm of the RED function used in the present invention is as follows. RED further increases the packet data discard rate to a value greater than 0 in accordance with an increase in the average buffer length obtained by averaging the buffer lengths over a certain period so as not to be affected by short-term fluctuations in the current buffer length. The average buffer length Qa is calculated according to Equation 1 from the current buffer length Q using exponential average (EWMA: Exponential Weighted Moving Average) for each arrival of packet data.

(Expression 1) Qa ← (1-Wq) Qa + WqQ
Here, Qa: average buffer length, Wq: parameter for determining the weight of the current buffer length (where 0 <Wq <1), Q: current buffer length.

  Then, two threshold values of the minimum threshold value MINth and the maximum threshold value MAXth are provided for the average buffer length Qa, and the packet data discarding process is performed. Specifically, as shown in FIG. 2, the packet data discard rate becomes MAXp when the average buffer length Qa starts to increase from MINth and the average buffer length Qa is MAXth. When the average buffer length Qa exceeds MAXth, all packets are discarded.

  The thresholds of these MINth, MAXth, and MAXp are RED parameters, and the degree of packet data discard is determined by these parameters. When the degree of discard is excessive, unnecessary RTO increases, and when it is insufficient, RTO increases due to buffer overflow. Therefore, in order to properly operate the RED function, it is necessary to maximize the effect of suppressing RTO by optimally setting the threshold value so as not to cause excessive or insufficient discarding of packet data.

  In the process of S2, if it is determined that the packet data input to the router is to be discarded by the RED function (S2; Yes), the discard flag management means of the router increases the value of the discard flag Df by 1 (S3).

  The discard flag Df is a variable for taking over a packet data discard command by the RED function for subsequent processing, and is set in units of routers or buffers.

  The discard flag Df takes one of 0, 1, and 2 and is set to 0 at the initial stage of the packet data discarding process, and becomes 1 when a discard command is issued by the RED function. Then, the router searches for packet data that satisfies the requirements of the control method of the present invention with the value of the discard flag Df being 1, that is, discards further instructions by the RED function before packet data that satisfies the requirements appears. In this case, the value of the discard flag Df becomes 2. Note that the maximum value of the discard flag Df is 2.

  On the other hand, when it is determined that the packet data is not discarded by the RED function (S2; No), the discard flag determination unit of the router confirms the value of the discard flag Df (S4).

  When the value of the discard flag Df is 0 (S4; No), the preferential count management unit of the router decreases the value of the preferential count Fc of the flow to which the packet data to be processed belongs by 1 (S7). Then, the router takes the packet data into the buffer (S13), and finishes the processing related to the packet data (END).

  The preferential count Fc is a variable for avoiding discard of a plurality of packet data in the same window, and is managed for each flow. The flow of packet data input to the router is identified by the address number in the IP header.

  The preferential count Fc is set to the initial preferential value N at the start of processing (START). The initial preferential value N is set to a range of about 1 to 60, preferably about 30 to 60, and most preferably about 50, but may be a value exceeding 60. The preferential count Fc may be zero or a negative value.

  Subsequent to the process of S3 or when the value of the discard flag Df is greater than 0 in the process of S4 (S4; Yes), the discard flag determination unit of the router confirms the value of the discard flag Df (S5).

  If the value of the discard flag Df is not 2 (S5; No), the preferential count determination unit of the router confirms the value of the preferential count Fc of the flow to which the packet data to be processed belongs (S6).

  When the value of the preferential count Fc is larger than 0 (S6; Yes), the preferential count management unit of the router decreases the value of the preferential count Fc of the flow to which the packet data to be processed belongs by 1 (S8). Then, the router takes the packet data into the buffer (S14), and finishes the processing related to the packet data (END).

  Thus, in the present invention, when the value of the flow preferential count Fc is larger than 0, packet data belonging to the flow is not discarded. As a result, the discarding of a plurality of packet data in the same window of the same flow is suppressed, and the occurrence of RTO is suppressed.

  Here, as shown in FIG. 3A, in the RED function, the packet data P1 to be discarded is simply selected probabilistically, and the packet data P5 to be discarded next is also simply selected probabilistically. That is, since the packet data is discarded without considering whether the already discarded packet data P1 and the next discarded packet data P5 are in the same window, the discarding of a plurality of packet data in the same window is suppressed. In other words, RTO occurs and the throughput performance of the network decreases.

  On the other hand, in the case of the present invention, a plurality of packet data in the same window is obtained by performing a preferential treatment so that a predetermined number of newly input packet data related to the flow to which the already discarded packet data P1 belongs is not discarded. Probabilistic avoidance of disposal.

  Specifically, as shown in FIG. 3B, the flow preferential count Fc is used to consider the flow preferentiality, so that the flow preferential count Fc to which the discarded packet data P1 belongs is greater than 0. If it is high, the packet data P5 selected to be discarded by the RED function is avoided, and subsequently input packet data is sequentially examined to find and discard the packet data P7 satisfying the discard condition. As a result, the discard of the packet data P5 in the same window as the window in which the packet data P1 has already been discarded is avoided, the occurrence of RTO is avoided, and the deterioration of the network throughput performance is prevented.

  Since the preferential treatment is intended for a flow in which packet data is discarded, only a short flow is not advantageous. In addition, since the preferential count Fc is consumed earlier as the flow with more packet data passing through the router, only the flow occupying a large band is not advantageous. In other words, in the present invention, only flows having specific characteristics are not extremely favored or suppressed, so that control of packet data transmission without bias is performed.

  Subsequently, when the value of the discard flag Df is 2 in the process of S5 (S5; Yes) and when the value of the preferential count Fc is 0 or less in the process of S6 (S6; No), the router resends. The packet data determination means determines whether the packet data to be processed is retransmitted packet data as an attribute of the packet data (S9).

  Whether or not the packet data to be processed is retransmitted packet data is identified by managing the sequence number of the TCP header of the packet data input to the router for each flow. The retransmission packet data management means manages the sequence number of the packet data for each flow.

  If the packet data to be processed is retransmitted packet data (S9; Yes), the router proceeds to the process of S8, and the preferential count management means preferentially counts Fc of the flow to which the packet data to be processed belongs. 1 is decreased by 1 (S8). Then, the router takes the packet data into the buffer (S14), and finishes the processing related to the packet data (END).

  As a result, discarding of packet data retransmitted from a flow in which packet data has already been discarded is avoided, occurrence of RTO is avoided, and deterioration of network throughput performance is prevented.

  On the other hand, if it is not retransmitted packet data (S9; No), the router discards the packet data without taking it into the buffer (S10).

  Subsequently, the discard flag management unit of the router decrements the value of the discard flag Df by 1 (S11), and the preferential count management unit sets the preferential count Fc of the flow to which the packet data to be processed belongs to the initial preferential value N (S12). Then, the router finishes the processing related to the packet data (END).

  Then, the router returns to the processing of S1, and repeats the processing from S1 to S14 for the packet data newly input to the router.

  Note that the value of the discard flag Df for each router or buffer and the value of the preferential count Fc for each flow are used as they are when processing the next input packet data. For this reason, the discard flag management means and the preferential count management means of the router return to the processing of S1 while storing these values. Further, the retransmission packet data management means returns to the process of S1 while storing the sequence number of the TCP header of the input packet data managed for each flow and used in the process of S9.

  Therefore, when the first discard command by the RED function is issued for the previously input packet data (S2; Yes), the discard flag Df = 1 is set (S3). If it is determined that the packet data is not to be discarded based on the preferentialness of the flow to which the packet data belongs (S6) and the attribute (S9) of the packet data, the next input is made with the discard flag Df = 1. Move on to packet data processing. Even if the discard command by the RED function is not issued for the next input packet data (S2; No), the discard flag Df = 1 and the discard command continuation state by the RED function, that is, discard of the packet data For packet data that is determined to be in a waiting state (S4; Yes) and for which no discard instruction has been issued by the RED function, the preferentialness of the flow to which the packet data belongs (S6) and the attribute (S9) of the packet data It is determined whether or not the packet data can be discarded. Also, when a discard command is issued by the RED function for the next input packet data in the state of the discard flag Df = 1 (S2; Yes), the discard flag Df = 2 is set (S3), and the flow It changes to the processing which does not consider the preferential property of (S5; Yes). As a result, all the packet data except the retransmitted packet data is discarded, so that the discarding of the packet data is not delayed, excessive suppression of the discarding of the packet data is prevented, and the buffer overflow is prevented.

  In addition, although the above-mentioned form is an example of the suitable form of this invention, it is not limited to this, A various deformation | transformation implementation is possible in the range which does not deviate from the summary of this invention. For example, in this embodiment, the discard control by the RED function, the discard control based on the preferential count Fc, and the discard control based on whether or not the retransmission is performed sequentially for each packet data input to the router. Not limited to this, all packet data input to the router while the discard flag Df changes from 1 to 2 is taken into the buffer, and the packet data having the smallest preferential count Fc in this period is selected from the buffer and discarded. You may make it do. In this case as well, since the dwell time in the buffer does not change, data transmission can be performed without causing a delay. As a result, the initial preferential value N, which is the initial value of the preferential count Fc, may be simply set to a sufficiently large value such as 99999, so that it is possible to save the trouble of setting according to the traffic situation of the network, and When discarding is instructed by the discard control by RED, any packet data is always selected and discarded regardless of the remaining status of the preferential count Fc, so that the performance of the present invention can be fully exhibited.

  In this embodiment, it is determined whether or not to discard the packet data in consideration of both the preferentialness (S6) of the flow to which the packet data belongs and the attribute (S9) of the packet data. However, the present invention is not limited to this, and only the preferentialness of the flow to which the packet data belongs may be considered. Also in this case, since the discarding of a plurality of packet data in the same window is suppressed as compared with the case where the packet data to be discarded is simply selected probabilistically by the RED function, the performance of the present invention can be exhibited. In this case, since a router having a layer 3 processing capability may be used, the cost can be reduced compared to the case of using a router having a layer 4 processing capability.

  In this embodiment, when the discard flag Df = 2 (S5; Yes), the flow data preferentiality (S6) is not considered, but the packet data attribute (S9) is considered. However, the flow preferentiality (S6) and the packet data attribute (S9) may not be considered. Specifically, in the case of Yes in the process of S5 of the present embodiment, the process of S10 may be performed next to discard the packet data as it is.

  Furthermore, in this embodiment, an example based on the premise that the Reno version is used as TCP has been mainly described. However, the TCP version to which the present invention is applicable is not limited to this, and other versions of TCP It is possible to apply to. The present invention is particularly effective when applied to TCP having a function of suppressing the data transmission amount when a plurality of packet data in the same window is discarded or when retransmitted packet data is discarded. .

  An embodiment of the router buffer management method of the present invention and router performance evaluation using the management method will be described with reference to FIGS.

  The performance evaluation in this example was performed by computer simulation. As the simulation software, OPNET Modeler version 11.5 of OPNET was used. 4, users U1, U2,..., U16 (hereinafter referred to as user Ux), servers S1, S2,..., S16 (hereinafter referred to as server Sx), user-side routers R0, A network composed of a server side router R1, a link between the user Ux and the user side router R0, a link L1 between the server Sx and the server side router R1, and a bottleneck line L2 between the user side router R0 and the server side router R1. Simulation was performed with the model.

  File transfer by FTP (File Transfer Protocol) is performed one-to-one between the user Ux and the server Sx. The server-side router R1 is mounted with the router buffer management method of the present invention. The user-side router R0 has only a function of referring to the destination IP address of the packet data and simply distributing the data to the destination user Ux, and does not perform packet data discard control or the like by the RED function. That is, in the network model of the present embodiment, the traffic from the server Sx to the user Ux overflows the buffer of the server side router R1, while the traffic from the user Ux to the server Sx is very small and the buffer of the user side router R0. Will not overflow.

  The link bandwidth of the link L1 is 100 Mbps, and the transmission delay is 1 ms. The link bandwidth of the bottleneck line L2 is 8 Mbps, and the transmission delay is 200 ms. Note that the link bandwidth of the bottleneck line L2 is set in consideration of making it small enough to cause congestion. In addition, the transmission delay of the bottleneck line L2 is set so that packet data does not enter the MAC layer buffer so that the influence of fluctuation absorption by the MAC (Media Access Control) layer buffer of the routers R0 and R1 can be ignored. This is set in consideration of compensating for the decrease in delay in the MAC layer buffer due to this by the transmission delay of the bottleneck line L2.

  The user Ux and the server Sx implement TCP Reno, the maximum window size is 65,535 bytes, and the packet size is 1,500 bytes.

  The RED function of the server side router R1 is based on the prior examination, so that the effect of suppressing the occurrence of RTO is maximized, the minimum buffer threshold MINth = 9 packets, the maximum buffer threshold MAXth = 46 packets, the maximum discard rate MAXp = 1 / 6. The case where the RED function of the server side router R1 has these conditions is called an optimum discard case.

  Of the flows between the user Ux and the server Sx (hereinafter referred to as the user Ux to the server Sx), the 14 flows of the user U3 to the server S3, the user U4 to the server S4,. In order to simulate the preceding traffic occupying the band first, an infinite length file is set to be transmitted constantly. That is, the bandwidth of the bottleneck line L2 is occupied in advance of 14 flows whose file length is not limited. In addition, after confirming that congestion occurs when the number of preceding flows is 7 or more as a result of prior examination, the number of flows is set to 14 in consideration of the number of flows sufficient to generate congestion.

  On the other hand, the two flows of the user U1 to the server S1 and the user U2 to the server S2 are set so that a fixed-length file of 300 Kbytes is transmitted at a cycle of 20 seconds as a simulation of additional traffic that enters two later. As a result, a state in which packet data overflows from the buffer of the server side router R1 and a congestion state occurs periodically is simulated. Although the file length of the additional traffic is not particularly limited, it is set to a length that generates the largest congestion based on prior examination. In addition, when the number of additional flows is 1 or more, congestion occurs. Therefore, the number of flows is set to 2 in consideration of the number of flows sufficient to generate congestion in the same manner as the preceding flow.

  In this model, the same result can be obtained even if the bandwidth, the number of flows, and the buffer size of the bottleneck line L2 are enlarged or reduced at the same magnification. Therefore, this network model can verify the effect of the present invention on a realistic large-scale network without being limited to the network scale of the present embodiment.

  The simulation time was 600 seconds. That is, 30 additional traffics occur during this time, and the network becomes congested each time.

  Here, it has been pointed out that the effect of the conventional RED is not stable because the optimum parameter setting differs depending on the traffic situation (S.Flod, R.Gummadi and S.Shenker; Adaptive RED: An (See Algorithm for Increasing the Robustness of RED's Active Queue Management, Technical Report, August 2001).

  Therefore, in this example, the suppression effect of the present invention against the increase in the RTO occurrence frequency due to the deviation from the optimum setting of the optimum disposal case was evaluated. Specifically, with the optimal discard case as a standard, the additional traffic is increased to 3 flows and congestion is increased to reduce the RED discard rate, and the preceding traffic is reduced to 10 flows. A simulation was also performed for an excessive discard case in which the RED discard rate was excessive by reducing congestion.

  Furthermore, as a setting of the buffer management method of the router of the present invention, a simulation was performed for each case where the initial preferential value N of the preferential count Fc was changed by 5 from 0 to 60. In the case of the initial preferential value N = 0, only the avoidance process for discarding retransmitted packet data is executed in the process of the present invention.

  Further, in order to verify the effect of the router buffer management method of the present invention, a case where only the RED function is implemented in the server side router R1 is set as a comparative example.

  In this embodiment, the ratio of the number of RTOs generated in one flow of congestion per one congestion is defined as the RTO occurrence rate, and the performance evaluation of the present invention is performed by the RTO occurrence rate reduction effect based on the comparative example. went. Note that the RTO occurrence rate was determined by averaging 30 times. Here, the default setting of the waiting time TCP when an RTO occurs is 3 seconds, and further increases when the degree of congestion is large. When an RTO occurs once, it exceeds 3 seconds. A waiting time is generated, and the data transmission is delayed correspondingly, thereby impairing the convenience of the user Ux.

  Using the above-described conditions, a simulation according to the value of the initial preferential value N according to the buffer management method of the router of the present invention and a simulation of a comparative example for each discard condition case (that is, an optimal discard case, an insufficient discard case, and an excessive discard case) And went. And the result shown in FIG. 5 was obtained as a transition with respect to the change of the initial preferential value N about the RTO occurrence rate according to the disposal condition case when the present invention is applied.

  In any disposal condition case, the RTO occurrence rate decreased rapidly with the increase of the initial preferential value N, and the change became moderate from around the initial preferential value N = 20. The initial preferential value N = 50 was the smallest value (second smallest value in the case of insufficient disposal). It should be noted that the effect of reducing the occurrence of RTO occurred not only in specific flows but also in all flows.

  On the other hand, the RTO occurrence rate of the comparative example was 9.5% in the optimum disposal case, 15.6% in the insufficient disposal case, and 15.2% in the excess disposal case.

  From these, the reduction effect of the RTO generation rate when the present invention is applied to the RTO generation rate of the comparative example is [(RTO generation rate of the comparative example) − (RTO generation rate of the present invention)] / (RTO generation of the comparative example) Rate) × 100 (%). In addition, the value in the case of the initial preferential value N = 50 was used for the RTO occurrence rate of the present invention. As a result, the reduction effect of the RTO occurrence rate according to the present invention was 88% in the optimum disposal case, 82% in the insufficient disposal case, and 78% in the excess disposal case.

  From the above results, it was confirmed that the use of the control method of the present invention greatly suppresses the occurrence of RTO compared to the comparative example which is a conventional control method.

  Moreover, since the initial preferential value N = 50 is the minimum value (the second smallest value in the case of insufficient waste) in any disposal condition case, the initial preferential value N is preferably set when using the present invention. It was confirmed that it should be about 30 to 60, most preferably about 50.

  In addition, the control method of the present invention exhibits stable performance in any disposal condition case, and no significant performance degradation is observed due to changes in the traffic situation of the network. It was confirmed that the influence of the situation change on the throughput performance was small. Further, it was confirmed that the problem of the conventional RED function that the effect is not stable because the optimum parameter setting differs depending on the traffic situation can be improved.

It is a flowchart explaining an example of embodiment of the buffer management method of the router of this invention. It is a figure which shows the relationship between the discard rate of packet data, and a threshold value. It is a conceptual diagram explaining avoidance of multiple packet data discard and retransmitted packet data within the same window according to the present invention. It is a figure explaining the structure of the simulation model of an Example. It is a figure which shows the result of the RTO incidence rate at the time of applying this invention in an Example. It is a block diagram which shows the structure of the network node in the control method of the conventional packet data transmission.

Explanation of symbols

U1, U2, ..., U16 users S1, S2, ..., S16 server L1 link L2 bottleneck line R0, R1 router

Claims (2)

  Of the packet data determined as discard candidates by the RED function of the router, the preferential count whose value decreases when it is managed for each flow to which the packet data belongs and is taken into the buffer of the router is retransmitted as not satisfying the discard condition. A buffer management method for a router, wherein the preferential count satisfies a discard condition and is not retransmitted from packet data subsequently input to the router without discarding the packet data . A RED executing means for determining whether or not to discard packet data by the RED function; a discard flag managing means for managing a discard flag whose value changes based on a determination result of the RED executing means; and a value of the discard flag. A discard flag determining unit that determines whether or not the packet data is a discard candidate based on the preferential count managing unit that manages a preferential count that decreases when the packet data is taken into the buffer for each flow; and the preferential count Based on preferential count determination means for determining whether or not the packet condition is satisfied, reproduction packet data management means for managing the sequence number of the packet data for each flow, and the sequence number managed for each flow Retransmission packet data determination means for determining whether or not the packet data has been retransmitted; Router which is characterized in that.

Images