JPH11355283A - Packet abandonment control method and node for realizing the method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a packet abandonment control method and a node for realizing the method capable of improving the throughput and relaxing the limit of the number of simultaneous connections by reducing a possibility to abandon consecutive TCP packets regardless of temporary congestion and reducing the uselessness of resources due to transfer of useless small-sized packets and reducing the abandonment of the small-sized packets due to the delay of window control at an originating side TCP. SOLUTION: When a small-sized packet is received, the queue length of a buffer which is regularly managed by a buffer monitor part and a previously determined threshold Tq are compared with each other by a queue comparison part. When the queue length exceeds the threshold Tg, a timer value is confirmed by a timer comparison part and in the case that this value exceeds a certain time, the certain number of the small-sized packets in the buffer are abandoned by a buffer control part and the timer value is initialized. In the case that the value does not exceed the threshold Tq, the small-sized packet is inputted without working anything on it.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention decomposes an information frame or a packet formed by decomposing an information frame (hereinafter referred to as an upper packet) into small packets,
The present invention relates to a packet discarding technique at the time of congestion of a node when transferring data by transferring the small packet.
(Local Area Network) A packet discarding control method which is effective for nodes constituting a network and which is effective for improving network throughput, and a node for realizing the method. About.

[0002]

2. Description of the Related Art FIG. 6 shows an outline of a typical conventional packet discarding method. In this method, the node always manages the queue length of cells stored in the buffer, and when the queue length becomes equal to the buffer length (physical capacity of the buffer) and the buffer becomes full, a new node is created. Is to discard small packets arriving at. That is, in the example shown in FIG. 6, three small packets (small packets 2, 3, and 4) are stored in a buffer having a buffer length of 4.
Is already stored (FIG. 6 (a)), and when two small packets (small packets 5, 6) arrive there, first, the small packet 5 is input and the buffer becomes full and the small packet 6 Because there is no room to store the
The small packet 6 is discarded (FIG. 6B).

Another conventional cell discarding method is shown in FIG.
(A. Romanow and S. Floyd,
“Dynamics of TCP Traffic over ATM Networks,” Proc.
ACMSIGCOMM Conference 1994, pp.79-88, 1994.).
In this method, the node always manages the queue length of the buffer, and when the queue length exceeds a certain threshold smaller than the buffer length, the small packet included in the small packet arriving and the upper packet to which the small packet belongs belongs. It is all to be discarded. That is, in the example shown in FIG. 7, for example, the threshold value Tq
Is set to 3 in advance. It is assumed that a small packet 2 of the upper packet A and small packets 3 and 4 of the upper packet B are stored in the buffer (FIG. 7A). When the upper packet C composed of the small packets 5 and 6 arrives there, the queue length has reached the threshold value, so that all the small packets included in the upper packet C including the small packet 5 are discarded (see FIG. 7 (b)).

A method shown in FIG. 8 has also been proposed by the same applicant as the present invention (see Japanese Patent Application No. 9-45203). In this method, the node always monitors the queue length of the buffer, and when the queue length exceeds a preset threshold value Tq, a certain number of small packets are continuously discarded from the buffer or the output side of the buffer. It was made. In the example of FIG. 8, a small packet length of 4 buffers and a small packet of 3 thresholds Tq (Tq =
In the case 3), three small packets (small packets 2, 3, and 4) are already stored. When the small packet 6 arrives, a fixed number (D = 2) of small packets 2 and 3 are discarded from the output side of the buffer because the queue length exceeds the threshold value (Tq = 3 packet lengths). The figure shows how small packets 7, 8, 9 arriving subsequently can be stored in the buffer.

[0005]

When the typical packet discarding method shown in FIG. 6 is used, the following problems occur at the time of congestion. 1) IP over A as an example of supported data transfer
A case will be considered in which TCP (Transmission Control Protocol) is used in an upper layer in TM (a method of transferring IP packets by ATM). In a window control method called fast-retransmit of TCP, the transmitting TCP assumes that congestion has occurred when receiving (normally three) Ack (Acknowledgment) requesting retransmission of a specific packet, and Make side window smaller (WRSteven
s, "TCP / IP Illustrated, Volume1," Addison Wesley, p
p.297-322, March 1996.).

In this case, the transmitting side recognizes the packet loss due to the buffer overflow, narrows the transmission window, and the time required for the queue length of the buffer to decrease due to the effect is the minimum Round Trip Time (RTT). Also, since this method discards ATM cells (small packets), the time required for the buffer to become congested again becomes short. Therefore, in this method, once congestion occurs, the minimum RT
Discrete cell (small packet) discards occur continuously many times during T, increasing the likelihood of discarding many consecutive TCP packets.

2) For example, when a method of dividing one upper-layer packet into a plurality of small packets of a lower layer and transmitting the packet, such as an ATM, is used in a typical packet discarding method. Since the packet is discarded in units of (small packets), a part of cells (small packets) constituting the upper layer packet is discarded. However, cells that have not been discarded (small packets) belonging to the same packet are also discarded after being transferred to the destination terminal. That is, cells (small packets) discarded at the destination terminal are transferred, and resources are wasted.

3) As an example, as in the above 1), IP ov
Consider a case where TCP is used in an upper layer of er ATM. In a typical packet discarding method, the arriving cell (small packet) at the time of congestion when the empty buffer becomes zero is discarded, so that the cells subsequent to the discarded cell (small packet) (small packet) pass through the buffer. We have to wait for a few minutes of cells (small packets) in the entire buffer. In this case, the next cell (small packet) following the discarded cell (small packet) for the waiting time
Arrives at the destination side late, and therefore the Ack is returned to the originating side too late. That is, the window control of the originating TCP is delayed. As a result, many cells (small packets) are discarded.

When many TCP packets are discarded within a limited short time, TCP fast-retransmission
Due to the nature of the above, only retransmission using the TCP timeout can be performed. Generally, the timer value of the TCP timeout is set to be much larger than the normal RTT, so that retransmission by this significantly lowers the throughput (K.Fall
and S. Floyd, “Simulation-based Comparisons of Taho
e, Reno, and SACK TCP, “Computer Communications Revi
ew, July 1996.). The conventional packet discarding method shown in FIG. 7 can slightly solve the above 1) and 2). However, this method cannot solve the above problem 3), but also has the following problem 4).

[0010] 4) There is a limit to the number of simultaneous connections because, for example, it is necessary to access a table describing whether to discard small packets or input them into a buffer. In the packet discarding method shown in FIG. 8, there is a high probability that the above 1), 2) and 3) can be alleviated. However, in this case, the interval at which the buffer is congested largely depends on the transmission rate on the transmitting side. Therefore, when the input rate is large and the interval is small, the following problem 5) occurs.

5) Consider a case where TCP is used in an upper layer. In this case, fast-retransmission caused by the sender
Of three duplicate Acs returned by the receiver that trigger
Before a buffer congested with three TCP packets corresponding to k does not output, the buffer may become congested again. In this case, fast-retran
Window control by smission is delayed, many TCP packets are discarded, and only retransmission by TCP timeout can be performed due to the inherent properties of TCP, and the throughput is greatly reduced.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above problem, to reduce the possibility of discarding continuous TCP packets even once congested, to reduce the waste of resources due to unnecessary small packet transfer, To provide a packet discarding control method capable of improving the throughput by reducing the discarding of small packets due to the delay of the window control and improving the limit of the number of simultaneous connections, and a node for realizing the method. is there.

[0013]

According to the present invention, in order to achieve the above object, a node always monitors a buffer (the length of B small packets), and the queue length of the buffer has a "threshold ( (Tq small packet lengths) ", a fixed number (D) of small packets are collectively discarded from the buffer. Alternatively, a fixed number (D) of small packets are collectively discarded from the output side of the buffer. After packet discarding, within a certain time (Tt: ms),
Even if the queue length exceeds the threshold value Tq, packet discard processing is not performed.

According to this configuration, since a large number of continuous small packets are discarded at one time, it is assumed that a network in which a large percentage of traffic flows higher than the capacity of the network due to a constant number of simultaneous connections. By doing so, the problems 1) and 2) described above can be alleviated and the problem 3) can be solved with very simple control. In addition, since it can be realized with very simple control, the problem described in 4) can be solved. Further, by setting the fixed time Tt to a value sufficient for the congestion buffer to output a sufficient number of higher-order packets for all connections using the buffer to cause the receiving side to recognize congestion, The problem shown in 5) can also be solved.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The configuration of the present invention will be described below in detail. As an example of the data communication to be applied, IPover AT
Description will be made on the assumption that a TCP packet such as M is carried by an ATM cell. FIG. 1 is a diagram for explaining storage / discarding of small packets in a buffer when the present invention is applied. The buffer length is 11 small packets, the threshold Tq is 7 small packets, and the small packets to be discarded. The case where the constant D is 3 is shown. FIG. 2 is a diagram illustrating an example of a network configuration assumed in the present embodiment. This is a model of the number of clients: the number of servers = a plurality (M units): a plurality (N units).
The figure shows a case in which M clients are connected via W1 and the receiving node ATM-SW2. The figure also shows the protocol structure used.

FIG. 3 shows a window size (Congestion Win) on the transmitting side when TCP is used in the upper layer as an example.
FIG. 9 is a diagram showing a transition model of dow: cwnd) (Mbps) and a transition of a queue length (Mbit) of a transmitting node corresponding to the transition model.

In FIG. 2, a server and a transmitting node AT
Assuming that the propagation delay between M-SW1 is Δt, in FIG.
T3, which is the time at which the window size cwnd of the transmitting side TCP becomes small due to the influence of the small packet discarding of T2, corresponds to T5, which is the time at which the queue length of the buffer actually starts to decrease. First, at time T1, the window exceeds the link band Vp, and the queue length of the buffer starts to increase due to the influence (see FIG. 1A). At T2, the queue length becomes equal to the threshold value Tq (FIG. 1B). reference),
Discard D small packets (cells) from the buffer.
Due to this discard, the queue length becomes BD (see FIG. 1C). Immediately after that, until T5 (= T2 + RTT), since the server does not recognize the packet discard, the queue continues to grow, and the queue length exceeds the threshold value Tq before T4 (= T2 + Tt) (FIG. 1 ( d)). However, in the present invention, since the timer for the fixed time Tt is running, small packets (cells) are not discarded when the threshold Tq is exceeded. T4 after a certain time Tt has elapsed from T2
At this point, the small packet (cell) is discarded again (see FIG. 1).
(E)).

At T3, the server TCP recognizes the packet discard at T2, retransmits the discarded TCP packet on the network, and reduces the window size. Due to this effect, the queue length starts to decrease at T5 (= T3 + Δt). T3
In T6 after the elapse of RTT, the Ack for the TCP packet retransmitted by the transmitting TCP can be received, and fast-retr
Exits the ansmission state.

Next, a description will be given of a configuration example of a node for realizing the present invention. FIG. 4 is a diagram illustrating a configuration example of a node that implements the present invention. In FIG. 1, reference numeral 1 denotes a buffer monitoring unit that constantly monitors the buffer queue length. 2
Is a comparison unit that compares the queue length of the buffer monitored in 1 with a threshold value. Reference numeral 3 denotes a timer unit that is initialized each time a small packet is discarded, and compares Tt with a timer value only when the queue length upon arrival of the small packet exceeds the threshold value Tq. Reference numeral 4 denotes a buffer control unit for discarding small packets from the buffer when the queue length exceeds the threshold value Tq as a result of the comparison at 2, and the timer value exceeds the threshold value Tt as a result of the comparison at 3. Is shown. 5 indicates a buffer of the node.

FIG. 5 is a flowchart showing a processing procedure in the node. Next, the processing on the node side will be described with reference to the configuration diagram of the node in FIG. 4 and the flowchart in FIG. First, when a small packet is received (step 10)
1) The queue comparing unit 2 compares the queue length constantly managed by the buffer monitoring unit 1 with the threshold value Tq (step 102). When the queue length exceeds the threshold value Tq (step 102: Y), the timer value is checked by the timer comparison unit 3, and when the value exceeds a predetermined time Tt (step 103: Y), the buffer control is performed. The unit 4 discards a fixed number (D) of small packets in the buffer and initializes the timer value (step 104). If the threshold value is not exceeded (step 103: N), input is made without any operation.

[0021]

As described above, according to the present invention,
In a congested state, there is a low possibility that small packets are continuously discarded on the input side, and the information packets constitute 1
It is unlikely that resources are wasted by carrying small packets discarded only at the terminal side and discarding small packets at the terminal side, and the time until the terminal takes control when the network becomes congested is shortened. . Therefore, the network can be used effectively, and the throughput of the network can be improved. Further, since complicated control such as extraction of a small packet belonging to a unique upper packet is not required, the problem of the limitation of the number of simultaneous connections, which is a drawback of the conventional technology shown in FIG. 2, is eliminated, and the feasibility is increased. .

[Brief description of the drawings]

FIG. 1 is a diagram showing an outline of a packet discarding method of the present invention.

FIG. 2 is a diagram showing a configuration example of a network to which the present invention is applied.

FIG. 3 is a diagram showing a transition model of a window size on the transmission side and a transition of a queue length of the transmission side node corresponding to the model.

FIG. 4 is a diagram illustrating an example of a node that executes the present invention.

FIG. 5 is a diagram showing a processing procedure on the node side according to the present invention.

FIG. 6 is a diagram for explaining a conventional typical packet discarding method.

FIG. 7 is a diagram for explaining an existing packet discarding method.

FIG. 8 is a diagram illustrating a previously proposed packet discarding method.

[Description of Signs] 1: Buffer monitoring unit, 2: Queue comparison unit, 3: Timer comparison unit, 4: Buffer control unit, 5: Buffer.

Claims (3)

[Claims] 1. A packet discard control method for a buffer in a node congestion in data communication in which an information frame or a packet formed by decomposing an information frame is decomposed into smaller packets having a smaller size and transferred. Monitoring the status and determining whether the queue length of the buffer exceeds a predetermined threshold; and determining in the determining step that the queue length of the buffer has exceeded the predetermined threshold. Having a step of collectively discarding a certain number of small packets from the buffer in the event of a failure, and a step of, after discarding the certain number of small packets, suppressing the discard of small packets even if the threshold value is exceeded within a certain period of time. A packet discarding control method. 2. The packet discard control method according to claim 1, wherein the step of discarding the fixed number of small packets from the buffer is performed from the output side of the buffer. 3. A node in data communication for decomposing an information frame or a packet formed by decomposing an information frame into smaller packets having a smaller size and transferring the packet, comprising: a buffer for holding a received small packet; Buffer monitoring means for monitoring the state of the buffer; comparison means for comparing the buffer queue length obtained by the buffer monitoring means with a predetermined threshold value; When a threshold value exceeds a predetermined threshold, a certain number of small packets in the buffer are discarded, and after a certain number of small packets are discarded, small packets are exceeded for a certain period of time even if the threshold value is exceeded. And a buffer control unit for suppressing discard of the data.