JP5661306B2 - Congestion control device and congestion control method - Google Patents

Description

  The present invention relates to a congestion control device and a congestion control method, and more specifically, when a buffer capacity of a node is exceeded, a congestion control device that returns a packet of the node to an upstream node to reduce congestion of the node. And a congestion control method.

In recent years, general-purpose communication devices based on low-cost TCP / IP have been applied not only to the Internet but also to closed networks in companies. And, as various applications are taken into the network, measures for preventing congestion are becoming more and more important.
The basic measure to prevent congestion is to increase network equipment so that congestion does not occur. However, if it is unavoidable to delay the increase for economic reasons, or redundancy / burden is required. When a failure occurs in a distributed communication path, there is a high possibility that congestion will occur at a node where traffic is concentrated. For this reason, it is necessary to take measures to efficiently perform congestion control even when congestion occurs.

Therefore, TCP, which is used as a transfer protocol in most applications, has a congestion control function that suppresses the communication speed in order to suppress traffic within the equipment capacity. That is, when congestion occurs in a node in the communication path and a packet is discarded, the TCP of the transmission source detects it and reduces the window size that determines the number of packets that can be transmitted at one time, thereby reducing the packet to the network. It suppresses the transmission speed and eliminates congestion.
As a conventional technique for resolving congestion, Patent Document 1 discloses that when a buffer of a node exceeds a predetermined threshold due to congestion, the packet is transferred to a common external buffer, temporarily held, and the congestion is settled. Discloses a technique for returning a packet to a node buffer. Also, in Patent Document 2, when a node buffer exceeds a predetermined threshold due to congestion, a control signal for reducing the speed is transmitted to a plurality of nodes, and the node that receives the signal reduces the speed and surplus. A method for eliminating congestion by temporarily holding a packet is disclosed. Patent Document 3 discloses a method of transferring a packet to a buffer for another port in the same node and temporarily storing it when the buffer of the node exceeds a predetermined threshold.

JP 2007-325178 A JP 2009-60283 A JP 2009-081595 A

  However, there is a problem in TCP congestion control. One of them is that when a packet is discarded, all the packets after the discarded packet are retransmitted from the transmission source, so that the transmission efficiency is lowered. As a conventional approach to this problem, there is Split-TCP (RFC2757). This is a method in which a TCP connection is reestablished at a specific node (called TCP proxy) in a communication path. That is, a pseudo confirmation response packet is returned to the transmission source as seen from the TCP proxy, and the reception packet is buffered by the TCP proxy and transmitted to the transmission destination. At this time, it is necessary to copy the packet. When the packet is discarded on the communication path ahead of the TCP proxy, the packet copied from the TCP proxy is retransmitted instead of the transmission source. As a result, the feedback loop of the TCP connection is reduced, and the transmission efficiency is improved. However, since Split-TCP places TCP proxies in a fixed manner, it is necessary to select an optimum placement position where an effect can be obtained. Further, there is a problem that even if the packet is not discarded, it is necessary to always copy the packet with the TCP proxy, and resources for that purpose are required.

Further, since the conventional technique disclosed in Patent Document 1 requires a common external buffer, there is a problem that an optimal arrangement position must be selected as in the case of the TCP proxy.
Further, in the prior art disclosed in Patent Document 2, a node in which congestion is occurring transmits a control signal for reducing the speed, and the node that receives the signal performs speed control. As a result, the node in which congestion is occurring Since feedback control is used to eliminate congestion, it is difficult to set the optimal buffer threshold and speed control according to the traffic situation. If this is not the optimal setting, there will be a problem of packet discard due to the control delay. To do.
The present invention has been made in view of such a problem. When congestion occurs in an arbitrary node in a communication path and the amount of received packets exceeds the buffer capacity of the node, packets exceeding the buffer capacity are processed. Improve the stability of congestion in a general IP communication network by performing processing to re-enter the buffer for returning to the upstream node as seen from that node, and returning the packet to the upstream node without discarding An object of the present invention is to provide a congestion control device and a congestion control method that can be used.

For the present invention to solve the above problems, according to claim 1 is provided in a specific node congestion occurs, the congestion controller to return the packet to the upstream node disposed upstream of the network than the specific node A buffer capacity detecting means for detecting that the buffer capacity of the specific node has been exceeded, and a packet selection for selecting a packet to be returned when the buffer capacity detecting means detects that the buffer capacity has been exceeded. means and a packet returning means for returning the packet selected by the packet selecting means to the upstream node, re-receiving means for re-receiving the retransmission packet retransmitted from the upstream node to the specific node on a per-flow basis When, and a control means for controlling said each means, said control means, the buffer capacity in the upstream node If there is no margin is characterized by returning the other packet to be the return to the upstream side node disposed upstream on the upstream node further the network than divided on a per-flow basis.
If there is no room in the buffer capacity in the upstream node, it is further divided and returned to the upstream node in units of flows. Similarly, when the communication path is branched and there are a plurality of upstream nodes, it is divided and returned in units of flow. Thereby, the buffer free capacity of a plurality of nodes can be utilized.

According to a second aspect of the present invention, when the control unit returns a packet exceeding the buffer capacity to the upstream node, the control unit returns the packet in a flow unit indicating a communication connection in which the transmission side and the reception side are associated one-to-one. It is characterized by that.
When returning a packet exceeding the buffer capacity to the upstream node, it is preferable to return the packet in units of flows. That is, it is assumed that packets that reach the buffer are packets of flows A, B, C, and D. When the packets of the flows A and B exceed the buffer capacity, the packets are returned to the upstream node, and thereafter, all the arrival packets are returned for the packets of the flows A and B. In this way, since the flows A and B do not enter the buffer, congestion is alleviated and the remaining flows C and D can pass through the buffer. As a result, it is possible to prevent the packet arrival order from being changed at the destination to which the packet finally arrives after re-transmission from the upstream node .

A third aspect of the present invention includes a buffer capacity detection unit, a packet selection unit, a packet return unit, a re- reception unit, and a control unit, and is provided in a specific node where congestion has occurred and is arranged upstream of the specific node on the network. A congestion control method of a congestion control apparatus for returning a packet to an upstream node , wherein the buffer capacity detecting means detects that the buffer capacity of the node has been exceeded, and the packet selecting means includes the buffer capacity when it is detected that the buffer capacity is exceeded by the detection means, and selecting to be sent back packet, the packet return means, and sending back a packet selected by the packet selecting means to the upstream node the re-reception means, said Japanese a retransmission packet is retransmitted from said upstream node per flow Receiving a node, the control means, and a step of controlling the respective means, wherein the step of controlling, when there is no room in the buffer capacity at the upstream node from the upstream node Further, the present invention is characterized in that the packet to be returned is divided and returned in units of flow to another upstream node arranged upstream on the network .
There exists an effect similar to Claim 1.

According to the present invention, a packet exceeding the buffer capacity is not discarded, but the process of re-entering the buffer for returning to the upstream node as viewed from the node is performed, and the packet is returned to the upstream node. Also, if the holding time exceeds a certain time, the source TCP retransmits the packet as a timeout, so the meaning of holding the packet is lost. For this reason, a retransmission time setting means for setting the retransmission time is required. As a result, even when congestion occurs in any node in the communication path and the amount of packets that have reached exceeds the buffer capacity of the node, congestion can be eliminated without discarding packets that exceed the buffer capacity. .
In addition, when returning a packet that exceeds the buffer capacity to the upstream node, it is returned in units of flow, so after re-sending from the upstream node, the packet at the destination where the packet finally arrives It is possible to prevent the arrival order from being changed.
Further, when there is not enough buffer capacity in the upstream node, it is further divided and returned to the upstream node in units of flows, so that the buffer free capacity of a plurality of nodes can be utilized.
Further, since the retransmission time set by the retransmission time setting means is controlled so as to be shifted for each flow, it is possible to prevent traffic at each node from increasing at once.

It is a block diagram showing the function of the congestion control apparatus which concerns on one Embodiment of this invention. It is a figure which shows an example of a network diagram. It is a flowchart explaining the packet return operation | movement of the congestion control apparatus 50 of this invention. It is a flowchart explaining the packet resending operation | movement of the congestion control apparatus 50 of this invention. (A) is a figure explaining the operation | movement returned to an upstream node, (b) is a figure explaining the operation | movement resent from an upstream node. (A) is a diagram for explaining an operation in the upstream node where the buffer capacity has no margin and is further divided and returned to the upstream node in units of flows, and (b) is an operation for retransmission from the upstream node. It is a figure explaining.

  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 block diagram showing functions of a congestion control device according to an embodiment of the present invention. The congestion control device 50 is installed in the core router 4 and is a congestion control device by returning a packet to the upstream node of the specific node where the congestion has occurred, and the buffer capacity for detecting that the buffer capacity of the specific node has been exceeded. When the detection means 5 and the buffer capacity detection means 5 detect that the buffer capacity has been exceeded, the packet selection means 7 for selecting a packet to be returned, and the packet selected by the packet selection means 7 as an upstream node The packet return means 6 for returning to the packet, the retransmission time setting means 9 for setting the retransmission time of the return packet held in the upstream node, and the upstream node based on the time set by the retransmission time setting means 9 A retransmission means 10 for retransmitting the returned packet to the downstream node, a control unit (control means) 8 for controlling each means, an Internet And it is configured to include an interface 11 for exchanging information with the bets, the. An edge router 2 is connected to the Internet 3.
When congestion occurs in an arbitrary node in the communication path and the amount of received packets exceeds the buffer capacity of the node, conventionally, the packet is discarded at that node. In the present embodiment, a packet exceeding the buffer capacity is not discarded, but a process of re-entering the buffer for returning to the upstream node as viewed from the node is performed, and the packet is returned to the upstream node. The returned packet is held in the upstream node and sent again to the downstream node. In addition, if the upstream node receives the return packet in units of flow and holds the packet as it is, there will be no packet arriving from the source and congestion can be avoided. However, the throughput of the held flow is temporary. Will be greatly reduced. Furthermore, when the holding time exceeds a certain time, the transmission source TCP retransmits the packet as a timeout, so the meaning of holding the packet is lost. For this reason, the retransmission time setting means 9 for setting the retransmission time is required. As a result, even when congestion occurs in any node in the communication path and the amount of packets that have reached exceeds the buffer capacity of the node, congestion can be eliminated without discarding packets that exceed the buffer capacity. .

  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 transmission 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 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 transmission data held by the transmission end 1 is pulled out to the buffer of the edge router 2. Based on this 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 a timeout.

  FIG. 3 is a flowchart for explaining the packet return operation of the congestion control device 50 of the present invention. The controller 8 uses the buffer capacity detection means 5 to verify whether congestion has occurred (Y in S1) and the buffer capacity of the node has been exceeded (S2). As a result of the verification, if the capacity does not exceed, the process returns to step S1. If the capacity exceeds in step S2 (Y in S2), the packet selection means 7 selects the packet to be returned, and the packet return means 6 returns the packet to the upstream node (S3). Whether the buffer capacity of the returned upstream node has a margin is verified (S4). If there is no margin, the node is returned to the upstream node (S6). If there is a margin in the buffer of the upstream node in step S4 (Y in S4), it is verified whether or not all the arrived packets have been returned (S5), and the process is repeated until the completion. If no congestion occurs in step S1 (N in S1), the packet is passed through the buffer (S7).

FIG. 4 is a flowchart for explaining the packet retransmission operation of the congestion control device 50 of the present invention. First, when retransmission is started (S10), the first packet held in the upstream node is retransmitted (S11). Next, in order to retransmit the packet for each flow, a predetermined time is shifted (S12), and the next packet is retransmitted (S13). Then, step S12 is repeated until retransmission of all packets is completed.
If congestion occurs at any node in the communication path and the amount of packets that arrived exceeds the buffer capacity of the node, the packet was discarded at that node. In the proposed method, the buffer capacity was exceeded. Instead of discarding the packet, the packet is returned to the upstream node as viewed from the node, and returned to the upstream node. The returned packet is held in the upstream node and sent again to the downstream node.
When a packet exceeding the buffer capacity is returned to the upstream node, it is returned in units of flow. A flow refers to a communication connection in which a transmission side and a reception side are connected one-to-one.

The return in units of flows will be specifically described with reference to FIG. FIG. 5A is a diagram for explaining the operation of returning to the upstream node, and FIG. 5B is a diagram for explaining the operation of retransmission from the upstream node. Assume that packets reaching the buffer are packets of flows A, B, C, and D. When the packets of the flows A and B exceed the buffer capacity at the node 21, the packets are returned to the upstream node 20, and thereafter, all the arrival packets are returned for the packets of the flows A and B. To do. In this way, since the flows A and B do not enter the buffer, the congestion is alleviated and the remaining flows C and D can pass through the buffer of the node 21. As a result, the packets of the flows A, B, C, and D are not returned little by little, but only the packets of the flows A and B are returned to the node 20.
Further, as shown in FIG. 5 (b), the packet is sent back to the upstream node by the flow unit, and after being retransmitted from the upstream node 20, the packet finally arrives at the transmission destination. (TCP can be rearranged at the destination even if the packet arrival order is changed in the middle, but if the order is changed more than a certain number, the source is (It is necessary to make sure that the order of the packets is not significantly changed.) Since the upstream node 20 is not congested, the buffer capacity is basically empty. The required free space for holding the returned packet is at most (number of returned flows) x (window size) (due to the nature of TCP, the transmission destination confirms the response to the transmission source every time a packet is received. Packets (hereinafter referred to as ACK) are returned, but the number of packets transmitted from the transmission source at a time by the time the ACK arrives is only the window size).

  FIG. 6 (a) is a diagram for explaining the operation of the upstream node having no buffer capacity, and further dividing the flow back to the upstream node in units of flows, and FIG. 6 (b) is a diagram from the upstream node. It is a figure explaining the operation | movement to resend. If there is no room in the buffer capacity in the upstream node, it is further divided and returned to the upstream node in units of flows. Similarly, when the communication path is branched and there are a plurality of upstream nodes 22 to 25, the communication paths are divided and returned in units of flow. By such divided return, the buffer free capacity of a plurality of nodes can be utilized. For example, it is assumed that 20% of the buffer capacity exceeds the capacity in a node where congestion has occurred. When a packet is returned to an upstream node and is divided and held by the four nodes 22 to 25, 5% of the buffer capacity of each node is used for packet holding (assuming that the nodes have the same buffer capacity). ).

Now, it is assumed that the upstream node receives and holds a return packet in units of flows. If the upstream node holds the packet as it is, there will be no packet that will eventually arrive from the transmission source, and congestion can be avoided, but the throughput of the held flow is temporarily greatly reduced. Furthermore, when the holding time exceeds a certain time, the transmission source TCP retransmits the packet as a timeout, so the meaning of holding the packet is lost. Therefore, the upstream node needs to immediately retransmit the returned packet to the downstream node. However, on the other hand, since there is a risk that congestion will occur again due to retransmission, it is desirable to prevent the traffic from increasing at once by, for example, processing for shifting the retransmission start time for each flow. In addition, after completing the transmission of the packet held in the upstream node, the packet is transmitted from the transmission source with the same window size as before the packet is discarded, so that it returns to the same speed as before the packet is discarded. It becomes. Although this speed recovery is quick, there is a high risk of falling into congestion again. Therefore, the packet for the next window size is also held for a certain holding time, and while holding time is gradually reduced, this is repeated several times with the window size, and the operation is performed to gradually increase the speed. Also good. Note that the speed can be adjusted more effectively when the TCP on the transmission side uses a method (such as TCP-Vegas) that adjusts the window size appropriately according to the change in RTT (Round Trip Time). Since the upstream node is a packet passage route, it is convenient to stably execute such processing.
Note that if the retransmitted packet overflows from the buffer again at the downstream node, it is highly possible that the TCP on the transmission side will retransmit as a timeout, so it is desirable to discard it as it is. Alternatively, a retransmitted packet may be preferentially passed through the buffer, and a packet that is not retransmitted may be returned to the upstream node.

DESCRIPTION OF SYMBOLS 1 Transmission end, 2 Edge router, 3 Internet, 4 Core router, 5 Buffer capacity detection means, 6 Packet return means, 7 Packet selection means, 8 Control part, 9 Retransmission time setting means, 10 Retransmission means, 50 Congestion control apparatus

Claims (3)

Provided to a particular node congestion occurs, a congestion controller to return the packet to the arranged upstream node upstream on the network than the specific node,
Buffer capacity detecting means for detecting that the buffer capacity of the specific node has exceeded,
A packet selecting means for selecting a packet to be returned when the buffer capacity detecting means detects that the buffer capacity has been exceeded;
A packet returning means for returning the packet selected by the packet selecting means to the upstream node,
And re-receiving means for re-receiving the retransmission packet retransmitted from the upstream node to the specific node on a per-flow basis,
Control means for controlling each of the means ,
When the buffer capacity is not sufficient in the upstream node, the control means sends the packet to be returned to another upstream node arranged further upstream on the network than the upstream node. A congestion control device that divides and returns the data. Wherein if to return the packets that exceed the buffer capacity to the upstream node, wherein, characterized in that return a flow unit refers to the communication connection that associates the sender and receiver in a one-to-one Item 2. The congestion control device according to Item 1. A buffer capacity detection unit, a packet selection unit, a packet return unit, a re- reception unit, and a control unit are provided in a specific node where congestion has occurred, and an upstream node disposed upstream of the specific node on the network. A congestion control method of a congestion control device for returning a packet ,
The buffer capacity detecting means detecting that the buffer capacity of the node has been exceeded;
Selecting the packet to be returned when the packet selecting means detects that the buffer capacity has been exceeded by the buffer capacity detecting means;
A step wherein the packet returning means for returning the packet selected by the packet selecting means to the upstream node,
A step wherein the re-reception means, for receiving the retransmission packet retransmitted from the upstream node to the specific node on a per-flow basis,
The control means comprising controlling the means ;
When the buffer capacity in the upstream node is not sufficient, the controlling step flows the packet to be returned to another upstream node arranged further upstream on the network than the upstream node. A congestion control method characterized by dividing and returning in units .

Images