JP2011061699A - Congestion control apparatus and congestion control method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a congestion control apparatus capable of reducing a packet discarding rate by enabling a behavior of an average buffer length to be predicted in early timing, and to provide a congestion control method. <P>SOLUTION: A congestion control apparatus 50 includes: a SYN/ACK detecting part 6 for detecting a packet, in response to a connection request from a connecting side, obtained by combining a connection permission from a connected side and the connection request; a mark detecting part 7 for detecting a predetermined mark included in a data packet; a packet/mark counting part 8 for counting the number of packets and the number of marks; a traffic variation-predicting part 10 for predicting variation of traffic based on the results counted by the packet/mark counting part 8, respectively; a RED parameter-varying part 11 for varying a parameter relating to random initial detection based on an amount of variation of the average buffer length predicted by the traffic variation predicting part 10; and a control part 9 for controlling the respective parts. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a congestion control device and a congestion control method, and more particularly to a congestion control device and a congestion control method for reducing a packet discard rate using traffic fluctuation prediction.

  A low-cost general-purpose communication device using TCP / IP communication (hereinafter simply referred to as TCP) has been applied not only to the Internet but also to private networks in companies. In addition, for increasing traffic, it is basic to increase the capacity of the network bandwidth, packet transfer throughput, etc., but when the increase is delayed or the line capacity is reduced due to a failure, the traffic is concentrated. Causes congestion. At that time, congestion control is performed in order to suppress the communication speed and suppress the traffic within the equipment capacity. Congestion control is performed by a TCP function used as a transfer protocol in most applications. TCP detects packet discard due to router buffer overflow due to network congestion at the sending end, and controls the packet transmission to the network by reducing the window size that determines the number of packets that can be transmitted at one time. Suppressed.

On the other hand, many routers currently employ the tail drop method for congestion control. However, in this method, packets are accumulated in the buffer in the order of arrival, and when the buffer is full, all subsequent packets are discarded, and therefore packet discarding occurs continuously during congestion due to traffic increase. At this time, by discarding packets over multiple flows, TCP synchronization that synchronizes and increases / decreases the window size, and RTO (Retransmission Time out), an operation that starts retransmission after waiting for a few seconds, There is a problem that the throughput is reduced.
In order to solve such a problem, an active queue management method has been conventionally proposed. A typical example is RED (Random Early Detection). RED is a method of probabilistically discarding arriving packets before the buffer is full in order to prevent packets from being continuously discarded due to buffer overflow. Thereby, the TCP packet transmission speed is controlled, and TCP synchronization or the like can be avoided. For this reason, the throughput can be improved as compared with the tail drop method. Also, RED is mounted on many commercially available routers and can be used or not.

However, RED also has problems. That is, RED increases the packet discard rate as the average buffer length increases as shown in FIG. The packet discard rate becomes MAXp when the average buffer length starts to increase from MINth and the average buffer length is MAXth. When the average buffer length exceeds MAXp, all packets are discarded. The threshold values of MINth, MAXth, and MAXp in the figure are RED parameters and determine the degree of packet discard. When this RED parameter is excessive, TCP synchronization and the like increase, and when it is insufficient, TCP synchronization and the like increase due to buffer overflow. Therefore, in order to sufficiently bring out the effect of improving the throughput of RED, the RED parameter must be set optimally. However, the optimal setting is very difficult because it varies depending on traffic fluctuations.
Therefore, ARED (Adaptive RED) or the like that changes the RED parameter in response to traffic fluctuations has been proposed. ARED etc. change the parameter of RED according to the change of average buffer length. However, since control is performed only after the average buffer length is detected, there is a limit to the tracking performance with respect to dynamic buffer length fluctuations.

Here, as a conventional technique related to congestion control, Patent Document 1 uses network feedback to predict when congestion is likely to occur in the future, and based on these predictions, end users in the system. A technology for suppressing congestion by changing the behavior of the system is disclosed.
In Patent Document 2, in order to predict traffic congestion, hyperlink information is extracted, the hyperlink information is aggregated to generate link information for each IP address, and traffic in a node is classified for each IP address. Measure and generate a traffic pattern. The traffic monitoring means monitors the traffic in the node for each IP address and compares it with the traffic pattern. It is disclosed that when the difference between the traffic pattern and the current traffic is equal to or greater than a threshold value, the possibility of congestion occurring is determined while referring to the link pattern.
In Patent Document 3, the flow control server predicts an increase or decrease in network traffic and issues a flow control instruction to the management server. The management server controls the routing information of each router based on the flow control instruction. It is disclosed that a routing control instruction is sent to each router, and the router sets its own routing information based on the routing control instruction.
In Patent Document 4, the transmission rate is appropriately changed according to the degree of occurrence of congestion, and the number of connections during transmission is measured. It is disclosed that a coefficient is determined based on the predicted number of connections and the actual number of connections in communication, and the next number of connections is predicted using this coefficient.

JP 2002-124991 A JP 2004-343466 A JP-A-2005-110079 JP 2005-245034 A

However, the prior art disclosed in Patent Document 1 delays a data packet transmitted by an end user until a time when the predicted congestion is alleviated in order to predict congestion. Is not necessarily highly accurate, and appropriate congestion control may not be performed.
In addition, the prior art disclosed in Patent Document 2 determines the possibility of congestion while referring to the link pattern when the difference between the traffic pattern and the current traffic is greater than or equal to the threshold. There is a problem that the timing of congestion control is delayed.
In addition, the related art disclosed in Patent Document 3 searches for information on traffic increase / decrease on the Internet and predicts traffic increase / decrease, and it is difficult to accurately search information on traffic increase / decrease. There's a problem.
The prior art disclosed in Patent Document 4 merely predicts the number of connections, and the accuracy of the prediction is not necessarily high, and there is a possibility that appropriate congestion control will not be performed.
The present invention has been made in view of such a problem, and detects a SYN / ACK transmitted from a transmission terminal or an edge router at the start of TCP, and adds a mark added to the header of the data packet before the FIN packet. It is configured to detect, by predicting the traffic fluctuation based on the number of detected SYN / ACK and marks, the behavior of the average buffer length is predicted at an early stage, and the RED parameter is changed, It is an object of the present invention to provide a congestion control device and a congestion control method that can sufficiently bring out the performance.

In order to solve such a problem, the present invention provides a congestion control apparatus in a TCP / IP communication network using traffic fluctuation prediction, which is connected to a connection request (SYN) from a connection side. A SYN / ACK detection unit that detects a packet that combines a connection permission (ACK) and a connection request (SYN) from the network, a mark detection unit that detects a predetermined mark included in the data packet from the connection side, and the SYN A packet mark counter for counting the number of packets detected by the / ACK detector and the number of marks detected by the mark detector, and traffic fluctuations based on the respective counting results counted by the packet mark counter Traffic fluctuation prediction unit to be predicted, and fluctuation amount of average buffer length predicted by the traffic fluctuation prediction unit Based characterized by comprising a RED parameter changer for changing the parameter of the random early detection, and a control unit for controlling the respective sections.
The present invention relates to a congestion control device and a congestion control method for reducing a packet discard rate by using traffic fluctuation prediction in a router. That is, the SYN / ACK packet sent from the transmitting end in the core router and the mark added to the data packet several packets before the end of communication are detected and counted, and the increase or decrease or difference of this count value is counted. In this invention, the RED parameter is changed based on the above, thereby reducing the packet discard rate due to RED. As a result, the behavior of the average buffer length can be predicted at an early stage, and the packet discard rate can be reduced.

According to a second aspect of the present invention, the mark is added to a data packet immediately before the end of communication on the connection side.
If the behavior of the average buffer length can be predicted earlier than before, the dynamic setting of the RED parameter can be started early, and the follow-up performance can be improved. Therefore, in the present invention, a mark is added to the data packet immediately before the end of communication on the connection side, and the mark is detected before the end of communication on the connected side. As a result, the dynamic setting of the RED parameter can be started from an early stage, and the follow-up performance can be improved.
According to a third aspect of the present invention, the control unit predicts that the average buffer length increases when the number of packets counted by the packet mark counting unit increases and the number of marks decreases, and relates to the random initial detection. The parameter is varied.
When the number of counted packets increases and the number of marks decreases, the window size is large. That is, since it is predicted that the average buffer length will increase, the RED parameter is changed to a value based on the average buffer length. Thereby, since the magnitude | size of average buffer length can be estimated at an early stage, appropriate follow-up performance can be realized.

The control unit predicts that the average buffer length rapidly increases when a difference between the number of packets counted by the packet mark counting unit and the number of marks exceeds a predetermined threshold, and the average The ratio of discarding packets when the buffer length reaches the maximum threshold is increased.
When the difference between the counted number of packets and the number of marks exceeds a predetermined threshold, it can be predicted that the average buffer length will increase rapidly. In such a case, when the average buffer length reaches the maximum threshold, the rate of discarding packets is increased, and packet discarding is promoted early to control congestion and prevent continuous discarding due to buffer overflow. It can be avoided.
According to a fifth aspect of the present invention, the control unit predicts that the average buffer length rapidly decreases when a difference between the number of packets counted by the packet mark counting unit and the number of marks is equal to or less than a predetermined threshold, and the average buffer The ratio of discarding packets when the length reaches the maximum threshold is reduced.
When the difference between the counted number of packets and the number of marks falls below a predetermined threshold, it can be predicted that the average buffer length will decrease rapidly. In such a case, excessive discards more than necessary can be avoided by reducing the rate of discarding packets when the average buffer length reaches the maximum threshold and suppressing packet discards early.

According to a sixth aspect of the present invention, the mark is at least a bit added to an option field in an IP header or a TCP header, or a URG bit added in a TCP head.
It is a necessary condition of the present invention that the mark added to the header part at the transmission end must have no influence on the TCP at the reception end, and is added as an example of a method for satisfying the requirement. The mark is a bit added to the option field in the IP header or TCP header, or a URG bit added in the TCP head. As a result, communication can be performed without affecting the TCP at the receiving end.
According to a seventh aspect of the present invention, when the router that aggregates a plurality of transmission end lines is an edge router and the other routers are core routers, the mark is added by the edge router instead of the transmission end. It is characterized by that.
In the cooperative operation of marking at the transmitting end (hereinafter referred to as a server) and reading the mark at the router, it is necessary to agree that each manufacturer of the server and the router adopts the same technical specifications. In general, this is likely to be difficult to realize due to the involvement of elements other than technical elements. Even if agreement is reached, it is good to introduce new servers and routers. However, when updating existing equipment sequentially, the more servers there are, the more the update timing of each server will be. Because of the difference, it is difficult to realize. Therefore, in the present invention, if a mark is attached by an edge router instead of a server, the core router reads the mark attached by the edge router, and the router manufacturer only adopts the technical specifications. Therefore, it is easy to implement.

Claim 8 is a congestion control method of a congestion control device comprising a SYN / ACK detection unit, a mark detection unit, a packet mark counting unit, a traffic fluctuation prediction unit, a RED parameter fluctuation unit, and a control unit, wherein the SYN A step of detecting a packet in which a connection request (SYN) from a connected side is combined with a connection request (SYN) in response to a connection request (SYN) from the connection side; A step of detecting a predetermined mark included in the data packet from the connection side, and a packet number counting unit detecting the number of packets detected by the SYN / ACK detection unit and the number of marks detected by the mark detection unit, respectively. Counting and the traffic fluctuation prediction unit based on the respective counting results counted by the packet mark counting unit. Predicting traffic fluctuations, changing the parameters related to random initial detection based on the fluctuation amount of the average buffer length predicted by the traffic fluctuation prediction part by the RED parameter fluctuation part, and the control part And a step of controlling each part.
The present invention has the same effect as that of the first aspect.

According to the present invention, the SYN / ACK packet sent from the transmitting end in the core router and the mark added to the data packet several packets before the end of communication are detected and counted, respectively. By changing the RED parameter based on the increase / decrease or difference, the packet discard rate due to RED can be reduced, so that the behavior of the average buffer length can be predicted at an early stage, and the packet discard rate can be reduced. Can be reduced.
In addition, since a mark is added to the data packet immediately before the end of communication on the connection side and the mark is detected before the end of communication on the connected side, the dynamic setting of the RED parameter can be started from an early stage. Performance can be improved.
Also, if the number of counted packets increases and the number of marks decreases, the window size is large, which is expected to increase the average buffer length. Therefore, the RED parameter is changed to a value based on the average buffer length. By doing so, the size of the average buffer length can be predicted at an early stage, and appropriate tracking performance can be realized.

In addition, when the difference between the counted number of packets and the number of marks increases beyond a predetermined threshold, the average buffer length can be predicted to increase rapidly, and the rate at which packets are discarded when the average buffer length reaches the maximum threshold is increased. Since the number of discarded packets is reduced, the number of discarded all packets can be suppressed even if the average buffer length increases rapidly.
When the difference between the counted number of packets and the number of marks falls below a predetermined threshold, the average buffer length can be predicted to decrease rapidly, and the rate of discarding packets when the average buffer length reaches the maximum threshold is reduced. Since the number of discarded packets is decreased, the number of discarded packets can be further suppressed when the average buffer length rapidly decreases.
Further, since the mark is a bit added to the option field in the IP header or TCP header or a URG bit added in the TCP head, communication can be performed without affecting the TCP at the receiving end.
If the edge router is marked instead of the server, it becomes a cooperative operation between routers in which the core router reads the mark added by the edge router, and it is only necessary to adopt the technical specifications only by the router manufacturer. Therefore, realization becomes easy.

It is a flowchart which shows the outline of the congestion control method which concerns on one Embodiment of this invention. It is a figure which shows an example of a network diagram. It is a sequence diagram which shows an example of the exchange of the packet performed between a transmission end and a receiving end. It is a block diagram explaining the function at the time of providing the congestion control apparatus of this invention in a core router. It is a flowchart explaining operation | movement of the congestion control apparatus of this invention. It is a figure which shows the relationship between average buffer length and a packet discard rate.

  Hereinafter, the present invention will be described in detail with reference to embodiments shown in the drawings. However, the components, types, combinations, shapes, relative arrangements, and the like described in this embodiment are merely illustrative examples and not intended to limit the scope of the present invention only unless otherwise specified. .

FIG. 1 is a flowchart showing an outline of a congestion control method according to an embodiment of the present invention. If the behavior of the average buffer length can be predicted earlier than before, the dynamic setting of the RED parameter can be started from an early stage, and the follow-up performance can be improved. For this reason, in the present invention, the behavior of the average buffer length is predicted as follows.
(1) When TCP communication is started, the window size is gradually opened at the transmission end and the communication speed is increased, which affects the increase in the buffer amount with a certain amount of time. Therefore, the router detects a SYN / ACK packet transmitted in the downstream direction when starting TCP communication (S1).
(2) During communication, the window size is gradually opened to increase the transmission speed (S2).
(3) When the TCP communication is terminated, the window size is closed at a stretch at the transmission end, which immediately affects the reduction of the buffer amount. In other words, in order to make a prediction in a certain amount of time, a device is necessary. For this reason, a mark for informing the end of the header of the data packet being transmitted is detected by the router immediately before the FIN packet transmitted in the downstream direction when the TCP communication is terminated (S3). That is, if the behavior of the average buffer length can be predicted early, the dynamic setting of the RED parameter can be started from an early stage, and the follow-up performance can be improved. Therefore, in the present embodiment, a mark is added to the data packet immediately before the end of communication on the connecting side (hereinafter referred to as the connecting side), and the mark is detected before the end of communication. As a result, the dynamic setting of the RED parameter can be started from an early stage, and the follow-up performance can be improved. In addition, the mark added to the header portion at the transmission end must have no influence on the TCP at the reception end. Therefore, in this embodiment, the mark is a bit added to the option field in the IP header or TCP header, or a URG bit added in the TCP head. As a result, communication can be performed without affecting the TCP at the receiving end.
(4) When the communication is completed, the window size is closed and the transmission speed is set to zero (S4).
(5) Finally, a FIN packet is transmitted and the process ends (S5).
(6) Count the number of packets and marks detected in step S1 and step S3, and predict traffic fluctuations based on the number (S6). For example, if the number of detected packets increases and the number of detected marks decreases, the average buffer length is expected to increase. In addition, the larger the difference in the number of detections, the more sudden increase is predicted. In order to improve the prediction accuracy, the window size, the bucket arrival interval, and the discard / retransmission packet may be managed for each flow. The flow can be identified from the transmission / reception IP address and the port number of the header portion, and the window size can be detected by the router by marking the header portion of the data packet that becomes the window size break at the transmission end.
(7) The RED parameter is changed based on the traffic fluctuation predicted in step S6 (S7). That is, MAXp in FIG. 6 is increased when traffic increases rapidly, and MAXp is decreased when traffic decreases rapidly.
(8) The packet is discarded based on the RED parameter (S8).

  It should be noted that the mark attached to the header part at the transmission end must be one that does not affect the TCP at the reception end. Therefore, for example, there are a bit added to an option field in an IP header or TCP header, and a URG bit added in a TCP header. The added mark may be removed after being recognized by the router. The function of marking the header portion at the transmission end needs to be implemented in a transmission terminal (such as a server), but it may become more difficult as the number of terminals increases. Therefore, a method of performing the cooperative operation between the routers and simply mounting on the routers will be described with reference to FIG.

FIG. 2 is a diagram illustrating an example of a network diagram. The router that aggregates the lines at the transmission end 1 is the edge router 2, and the other router is the core router 4. A plurality of edge routers 2 exist in the network 3. When the edge router 2 receives the packet from the transmission end 1, it immediately returns an acknowledgment (ACK) packet to the transmission end 1. On the other hand, the packet received by the edge router 2 is sent to the core router 4, and upon arrival at the receiving end, an acknowledgment (ACK) packet is received from the receiving end. That is, the TCP connection is re-established by the edge router 2. At this time, when viewed from the edge router 2, the network on the transmitting end 1 side has a small RTT (Round Trip Time), so the communication speed is fast. Conversely, the network on the core router side has a high RTT (particularly during congestion), so the communication speed is high. slow. Therefore, packets are naturally accumulated in the buffer of the edge router 2 due to this speed difference. This is an image in which the transmission data held by the transmission end 1 is pulled out to the buffer of the edge router 2. Based on such a principle, no communication delay occurs even if packets are accumulated in the buffer. In addition, since this buffer does not accumulate packets connected by a TCP connection, even if the holding time is long, TCP is not disconnected due to timeout. The buffer is used to mark a packet one step before the FIN packet arrives (how much is before the set value). Therefore, there is no need to accumulate more packets, and the edge router 2 performs control to match the communication speeds on the transmitting end 1 side and the core router 4 side so that excessive packets are not accumulated in the buffer.
The required amount of the buffer of the edge router 2 is basically (the number of packets before the FIN packet arrives (setting value)) × (the number of simultaneous connections). Control of the communication speed on the transmitting end 1 side by the edge router 2 is performed by adjusting the reply timing of the acknowledgment (ACK) packet to the transmitting side or increasing / decreasing the notification of the amount of packets that can be received at one time. .

The edge router 2 has a function of marking a header portion of a packet to be transmitted to the core router 4, and the mark is read by the core router 4 and used for RED control. Such an operation may be performed only when congestion occurs. When congestion occurs in the core router 4, the packet is discarded by RED, and the edge router 2 detects this to reduce the window size and lower the communication speed on the core router 4 side. At the same time, control for reducing the communication speed on the transmission end side is performed.
That is, in the cooperative operation of marking at the transmitting end (hereinafter referred to as a server) and reading the mark at the router, it is necessary to agree that each manufacturer of the server and the router adopts the same technical specifications. In general, this is likely to be difficult to realize due to the involvement of elements other than technical elements. Even if agreement is reached, it is good to introduce new servers and routers. However, when updating existing equipment sequentially, the more servers there are, the more the update timing of each server will be. Because of the difference, it is difficult to realize. Therefore, in the present invention, if the mark is added by the edge router 2 instead of the server, the core router 4 reads the mark added by the edge router 2, and the inter-router cooperative operation is performed. Since it becomes better to adopt, it is easy to realize.

  FIG. 3 is a sequence diagram showing an example of packet exchange performed between the transmitting end and the receiving end. In this example, the transmission end is described as an edge router and the reception end is described as a core router. First, a connection request (SYN) is transmitted from a receiving end which is a connection side to a transmission end which is a connected side (hereinafter referred to as a connected side). The transmitting end that has received SYN returns a packet that combines connection permission (ACK) and connection request (SYN) to the receiving end. The receiving end detects SYN / ACK and returns a connection permission (ACK) to the transmitting end. This is called a three-way handshake and a connection is established. As a result, the transmission end can transmit data. Here, in TCP, in order to communicate efficiently, the receiving end advertises the window size that can be received, but instead of transmitting data at the window size suddenly, the amount of data below the window size is initially set. Send. By increasing the number of segments to be transmitted each time an ACK is received, the bandwidth can be effectively used. The transmitting end adds a mark M to a data packet several packets before the end of transmission. The mark M is detected at the receiving end. Finally, the transmitting end sends a FIN packet informing the end of communication to the receiving end, the receiving end sends ACK / FIN to the transmitting end, and receives the ACK from the transmitting end, thereby completing the communication. The above describes one flow between the one-to-one transmission end and reception end. However, since packets arrive at the core router in FIG. 2 randomly from a plurality of edge routers, SYN / ACK and mark M are detected. Each time, the number is counted, and the traffic fluctuation is predicted based on the counted number.

FIG. 4 is a block diagram for explaining functions when the congestion control apparatus of the present invention is provided in the core router. The congestion control device 50 includes a SYN / ACK detection unit 6 that detects a packet that combines a connection request (SYN) and a connection request (SYN) from a connected side with respect to a connection request (SYN) from the connection side, A mark detection unit 7 for detecting a predetermined mark included in a data packet from the connection side, and a packet mark count for counting the number of packets detected by the SYN / ACK detection unit 6 and the number of marks detected by the mark detection unit 7 Unit 8, a traffic fluctuation prediction unit 10 that predicts traffic fluctuations based on the respective counting results counted by the packet mark counting part 8, and a fluctuation amount of the average buffer length predicted by the traffic fluctuation prediction part 10 A RED parameter changing unit 11 for changing a parameter related to random initial detection, and a control unit 9 for controlling each unit It is configured to include an interface 5 for connecting the core router 4 and the Internet 3.
The present embodiment is an invention relating to a congestion control device and a congestion control method for reducing a packet discard rate by using traffic fluctuation prediction in a router. That is, the SYN / ACK packet sent from the transmission end in the core router 4 and the mark added to the data packet several packets before the end of communication are detected and counted, and the count value is increased or decreased. By changing the RED parameter based on the difference, the packet discard rate due to RED is reduced. As a result, the behavior of the average buffer length can be predicted at an early stage, and the packet discard rate can be reduced.

  FIG. 5 is a flowchart for explaining the operation of the congestion control apparatus of the present invention. When communication is started, a connection request (SYN) is transmitted from the receiving end on the connection side to the transmission end on the connected side. The transmitting end that has received SYN returns a packet that combines connection permission (ACK) and connection request (SYN) to the receiving end. The packet is detected by the SYN / ACK detector 6 via the interface 5 (S10). Thus, SYN / ACK is counted by the packet mark counting unit 8 (S11). Thereafter, data transmission is started. The transmitting end adds a mark M to a data packet several packets before the end of transmission. The mark M is detected by the mark detection unit 7 (S12), whereby the mark M is counted by the packet mark counting unit 8 (S13). Finally, since the sending end sends a FIN packet notifying the end of communication to the receiving end, when receiving the FIN packet (Y in step S14), the control unit 9 determines the SYN from each count result counted by the packet mark counting unit 8. When the / ACK packet increases and the mark M decreases (Y in step S15), the average buffer length is predicted to increase (S16), and the RED parameter is changed based on the predicted value (S17). A packet by RED corresponding to the fluctuation amount is discarded (S18). If the SYN / ACK packet increases and the mark M does not decrease in step S15 (N in step S15), and the difference between the SYN / ACK packet and the mark M is large (Y in step S19), the average buffer length Is increased rapidly (S20), and MAXp in FIG. 6 is increased (S21). A packet by RED corresponding to the fluctuation amount is discarded (S18). If the difference between the SYN / ACK packet and the mark M is not large in step S19 (N in step S19), it is predicted that the average buffer length will decrease rapidly (S22), and MAXp in FIG. 6 is decreased (S23). . A packet by RED corresponding to the fluctuation amount is discarded (S18).

When the number of counted packets increases and the number of marks decreases, the window size is large. That is, since it is predicted that the average buffer length will increase, the RED parameter is changed to a value based on the average buffer length. Thereby, since the magnitude | size of average buffer length can be estimated at an early stage, appropriate follow-up performance can be realized.
In addition, when the difference between the counted number of packets and the number of marks exceeds a predetermined threshold value, the average buffer length can be predicted to increase rapidly. In such a case, when the average buffer length reaches the maximum threshold, the rate of discarding packets is increased to reduce the number of discarded packets. As a result, even if the average buffer length increases rapidly, discarding of all packets can be suppressed.
Further, when the difference between the counted number of packets and the number of marks is equal to or less than a predetermined threshold, it can be predicted that the average buffer length will rapidly decrease. In such a case, the ratio of discarding packets when the average buffer length reaches the maximum threshold is reduced to reduce the number of discarded packets. As a result, when the average buffer length rapidly decreases, it is possible to further suppress discarding of packets.

  1 sending end, 2 edge router, 3 network, 4 core router, 5 interface, 6 SYN / ACK detection unit, 7 mark detection unit, 8 packet mark counting unit, 9 control unit, 10 traffic fluctuation prediction unit, 11 RED parameter fluctuation , 50 Congestion Control Device

Claims (8)

A congestion control apparatus in a TCP / IP communication network using traffic fluctuation prediction,
A SYN / ACK detection unit for detecting a packet in which a connection request from the connected side and a connection request are combined in response to a connection request from the connection side;
A mark detection unit for detecting a predetermined mark included in the data packet from the connection side;
A packet mark counter for counting the number of packets detected by the SYN / ACK detector and the number of marks detected by the mark detector;
A traffic fluctuation prediction unit that predicts traffic fluctuations based on the respective counting results counted by the packet mark counting unit;
A RED parameter variation unit that varies a parameter related to random initial detection based on a variation amount of an average buffer length predicted by the traffic variation prediction unit;
A control unit for controlling each unit;
A congestion control apparatus comprising:   The congestion control apparatus according to claim 1, wherein the mark is added to a data packet immediately before the end of communication on the connection side.   When the number of packets counted by the packet / mark counting unit increases and the number of marks decreases, the control unit predicts that the average buffer length will increase and varies the parameter for the random initial detection. The congestion control device according to claim 1, wherein   The control unit predicts that the average buffer length rapidly increases when the difference between the number of packets counted by the packet mark counting unit and the number of marks exceeds a predetermined threshold, and the average buffer length is a maximum threshold. The congestion control apparatus according to claim 1, wherein a rate of discarding packets when the threshold is reached is increased.   The control unit predicts that the average buffer length suddenly decreases when the difference between the number of packets counted by the packet mark counting unit and the number of marks is equal to or less than a predetermined threshold, and the average buffer length becomes the maximum threshold. The congestion control apparatus according to claim 1, wherein a rate of discarding the packet when it reaches is reduced.   The congestion control according to any one of claims 1 to 5, wherein the mark is at least a bit added to an option field in an IP header or a TCP header, or a URG bit added in a TCP head. apparatus.   The router is configured so that the mark is added by the edge router instead of a server when a router that aggregates a plurality of transmission end lines is an edge router and the other routers are core routers. Item 7. The congestion control device according to any one of Items 1 to 6. A congestion control method for a congestion control device comprising a SYN / ACK detection unit, a mark detection unit, a packet mark counting unit, a traffic variation prediction unit, a RED parameter variation unit, and a control unit,
The SYN / ACK detection unit detecting a packet that combines a connection request from a connected side and a connection request with respect to a connection request from a connection side;
The mark detection unit detecting a predetermined mark included in a data packet from the connection side;
The packet mark counting unit counting the number of packets detected by the SYN / ACK detection unit and the number of marks detected by the mark detection unit;
The traffic fluctuation prediction unit predicting traffic fluctuations based on respective counting results counted by the packet mark counting unit;
The RED parameter changing unit changing a parameter related to random initial detection based on a fluctuation amount of an average buffer length predicted by the traffic fluctuation predicting unit;
The control unit controlling each unit;
A congestion control method for a congestion control apparatus, comprising:

Images