KR100482687B1 - Congestion Control Apparatus And Method For UBR Service In ATM Switch - Google Patents

Abstract

The present invention provides a congestion control for allocating a buffer for each cell constituting a packet in an Asynchronous Transfer Mode (ATM) switch and storing or discarding each cell according to the state of the VC buffer. The present invention relates to a congestion control apparatus and method for UBR service in an ATM switch. In the past, when TCP congestion occurs due to an overload of an ATM network, a large delay is caused in a TCP flow, causing a global synchronization. Since it affects the connection and has unfairness on bursty traffic and relatively long packets, there is a sudden drop in network traffic, and during this time, the network has abnormal transmission efficiency.

Accordingly, the present invention determines the possibility of congestion on the VC buffer of the ATM switch, and discards the currently delivered cell according to the packet discard probability, thereby improving the efficiency of the buffer and thus inducing the possibility of full synchronization even in a plurality of TCP connections. VC queuing to support FIFO services by allocating buffers for each cell for each VC, and outputting each cell in a round-robin fashion at the output of the corresponding VC buffer, This can improve unfairness on long packets.

Description

Congestion Control Apparatus And Method For UBR Service In ATM Switch}

The present invention relates to congestion control for Unspecified Bit Rate (UBR) service in an Asynchronous Transfer Mode (ATM) switch, and in particular, allocates a buffer for each virtual channel (VC) for each cell constituting a packet, An apparatus and method for congestion control for UBR service in an ATM switch configured to perform congestion control for storing or discarding each cell according to a state.

In general, ATM allows a wide variety of traffic such as data, voice, and video to be transmitted through a single network. The traffic types include different maximum delays, delay variations, It has characteristics such as timing, congestion control, and the like.

Ideally, an ATM switch should support performance that satisfies all Quality of Service (QoS). Among them, service for Unspecified Bit Rate (UBR) traffic uses a predetermined useful bandwidth to reduce cell delay. As an advantageous service for insensitive traffic transmission, UBR service provided in ATM network does not support explicit congestion control scheme.

For this reason, the conventional UBR service supports the flow control scheme of the Transmission Control Protocol (TCP) protocol, which supports data loss recovery and retransmission schemes at a higher level. It does not satisfy the quality.

In addition, when using a buffer switch limited to a predetermined size for the above-described UBR service, if the ATM-UBR network becomes overloaded and any cells constituting the packet are discarded, the effective output rate of the packet through the ATM network is very low. However, there is a serious problem in the fairness of bandwidth distribution for each TCP connection, and wasteful network resources are consumed in terms of bandwidth and buffer since unusable cells of damaged packets are continuously transmitted over the congested link.

At this time, the AAL type 5 (ATM Adaptation Layer Type 5) protocol applied to the UBR service implements a simple header structure to enable fast and efficient data communication, and the AAL type 5 cell has a payload type (PT; Payload). The AAL_indicate bit in the Type field indicates the start and end of the Common Part Convergence Sublayer-Protocol Data Unit (CPCS-PDU). The start and middle cells of the CPCS-PDU have an AAL indication bit. 0 'is set and transmitted, and the last cell is transmitted with the AAL indication bit set to' 1 ', thereby allowing the end user to distinguish the boundary of the packet.

On the other hand, in the past, damaged packets caused by overload in ATM networks cause network resources such as bandwidth and buffers to be wasted. Therefore, some congestion control schemes have been released to improve this, and a representative example thereof is PPD (Partial Packet Discarding). And EPD (Early Packet Discarding).

However, the PPD method discards all other cell streams, leaving only the last cell constituting the packet (AAL_indicate = 1) when the buffer overflows, which does not deal with cells of corrupted packets already written to the buffer. There is a disadvantage in that it is transmitted without, and the receiving side again requests for retransmission, causing waste of bandwidth and buffer.

In addition, the EPD method sets a predetermined threshold in the buffer and discards all cell streams of newly arrived packets when the number of cells in the buffer exceeds the threshold. Cells of corrupted packets written to the buffer have a disadvantage of causing a waste of bandwidth and buffer as they are transmitted to the end user.

It also supports TCP congestion control algorithms that control cell flow at higher levels, which control cell flow responsively to a traffic congested network while repeating the slow-start phase, congestion avoidance phase, and retransmission phase. .

Here, in the slow-start phase, the window is started by one packet while the sender increases the window up to a predefined congestion window until the response signal is received, and thus the window at each round trip time. Doubling the transmission.

Then, in the congestion avoidance state, when the window arrives at the maximum congestion window, the window is increased by one packet each time the response signal is received, and the congestion occurs and the response signal is transmitted to the sender within the timeout period. If they do not arrive, they are considered packet loss

When the packet loss occurs, it is responsive to the traffic congestion network by transitioning to the retransmission step, resetting the TCP congestion window to '1', and then transitioning back to the first step, the slow-start phase. To control.

However, conventionally, as the ATM network is congested due to overload, the lost packets have to be retransmitted, thereby increasing the load of the network, and causing global delay while incurring a significant delay in the TCP flow. This will affect many TCP connections.

Thus, as many TCP connections enter and exit the slow-start phase at about the same time, another burst occurs, and the cycle of congestion and depletion is repeated, for bursty traffic and for relatively long packets. Since there is an unfairness, there is a problem that a sudden drop in network traffic occurs, and the network has an abnormal transmission efficiency during this time.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and its object is to determine the possibility of congestion on the VC buffer of the ATM switch and to improve the efficiency of the buffer by discarding the currently delivered cell according to the packet discard probability. This eliminates the possibility of causing full synchronization even with multiple TCP connections.

Another object of the present invention is to provide VC queuing for supporting FIFO services by allocating buffers for each cell for each VC, and outputting each cell in a round-robin manner at an output of the corresponding VC buffer. It is to improve the unfairness of traffic and relatively long packets.

A feature of the present invention for achieving the above object is a cell classifier for classifying each cell of the TCP packet coming into the input link for each VC; An input cell counter for counting and reporting cells delivered for each VC from the cell classifier; A plurality of VC buffers for temporarily storing cells input for each VC in a corresponding buffer; A buffer monitoring unit for monitoring and reporting a state of each VC buffer; Each cell classified by VC by the cell classifier is stored or discarded in a corresponding VC buffer, and according to a cell count value reported by the input cell counter and each VC buffer status information reported by the buffer monitoring unit. A switch module for storing or discarding a currently input cell in a corresponding VC buffer; Congestion control for UBI services in an AT switch comprising a cell service scheduler for outputting each cell to an output link based on a dynamic round-robin scheme for periodically allocating cell slots for cells stored in the VC buffers To provide a device.

The switch module may be configured to perform VC queuing for allocating a logical buffer for each VC and storing each cell in a FIFO manner.

Another aspect of the present invention includes the steps of classifying each cell constituting the TCP packet input through the input link for each VC; Checking whether the currently input cell is the first cell of the packet by examining the AAL indication bits of each cell classified by VC; Checking whether the average queue size of the VC buffer is smaller than the average queue threshold when the currently input cell is the first cell of the packet; If the average queue size is smaller than the average queue threshold and the current queue size is smaller than the maximum threshold, storing the currently input cell in a corresponding VC buffer; When the average queue size is smaller than the average queue threshold and the current queue size is larger than the maximum threshold, it provides a congestion control method for the UBI service in the AMT switch, which includes discarding a currently input cell.

Checking whether the currently input cell is not the first cell of the packet as a result of examining the AAL indication bits of the cells classified for each VC; In the case of a cell of a damaged packet, it further comprises the step of discarding the currently delivered cell.

The method may further include discarding the first cell that is currently input when the average queue size of the VC buffer is greater than the average queue threshold and the current queue size is greater than the maximum threshold when the currently input cell is the first cell of the packet. It is done.

Hereinafter, an embodiment according to the present invention will be described in detail with reference to the accompanying drawings.

In the ATM switch according to the present invention, the structure for congestion control for UBR service is as shown in FIG. 1. The cell descriptor 11, the input cell counter 12, and the switch module 13), a plurality of VC (Virtual Channel) buffers 14-1 to 14-n, a buffer monitoring unit 15 and a cell service scheduler 16.

The cell classifier 11 classifies each cell of the TCP packet coming into the input link by VC to the switch module 13, and the input cell counter 12 inputs the cell classifier 11 for each VC. The counted cells are counted and reported to the switch module 13.

The switch module 13 stores or discards the cells classified for each VC by the cell classifier 11 in the corresponding VC buffers 14-1 to 14-n, and reports them by the input cell counter 12. According to the cell count value and the VC buffer status information reported by the buffer monitoring unit 15, it is determined whether to store or discard the cell currently input in the corresponding VC buffers 14-1 to 14-n. In addition, VC queuing (Per-VC queuing) for storing cells in each VC buffer 14-1 to 14-n supports a FIFO (First In First Out) service.

Each of the VC buffers 14-1 to 14-n temporarily stores a cell stored by the switch module 13, and the corresponding buffer monitoring unit 15 stores each of the VC buffers 14-1 to 14-n. Status is monitored and reported to switch module 13.

The Cell Service Scheduler (16) is based on a dynamic round-robin scheme that periodically allocates cell slots for cells stored in each of the VC buffers 14-1 to 14-n. Output each cell to the output link.

The congestion control operation for the UBR service in the ATM switch according to the present invention having such a structure will be described with reference to FIG. 2 as follows.

First, the VC buffers 14-1 to 14-n logically allocated to each VC are based on an output buffering model as shown in FIG. 3 and a cell slot is periodically determined by the cell service scheduler 16. Cells are output to the output link based on a dynamic round-robin scheme, in which each VC buffer 14-1 to 14-n has separate queuing to the switch module 13 , Hatched areas represent cells stored in the respective VC buffers 14-1 to 14-n.

The average queue threshold THavg is set between the minimum threshold THmin and the maximum threshold THmax for the VC buffers 14-1 to 14-n to increase the efficiency of the buffer. The average size of the total cells stored in 1 to 14-n) is defined as the average queue size (Qavg).

At this time, the average queue size and the size of each threshold have the order of 'maximum threshold-average queue threshold-average queue size-minimum threshold'.

When a predetermined packet is input through the input link in the defined state as described above (step S21), the cell classifier 11 classifies each cell constituting the packet by VC and then switches the module. (Step S22), at this time, the input cell counter 12 counts the cells input for each VC from the cell classifier 11 and reports them to the switch module 13.

Accordingly, the switch module 13 stores or discards the cells classified by VC in the corresponding VC buffers 14-1 to 14-n as they are delivered. To this end, the payload of the currently delivered cell is first used. It is checked whether the first cell of the packet is checked by checking whether the AAL indication bit in the type field is '0' or '1' (step S23).

If the currently delivered cell is identified as the first cell of the packet, the switch module 13 checks whether the average queue size of the VC buffers 14-1 to 14-n is smaller than the average queue threshold (step S24). If the average queue size is smaller than the average queue threshold, it is checked whether the current queue size is smaller than the maximum threshold (step S25).

At this time, when it is determined that the current queue size is smaller than the maximum threshold, the first cell transferred from the cell classifier 11 is stored in the corresponding VC buffers 14-1 to 14-n (step S26). If the queue size is larger than the maximum threshold, the first cell delivered from the cell classifier 11 is discarded (step S28).

In step S24, if the average queue size is larger than the average threshold, it is checked whether the current queue size is larger than the maximum threshold (step S27). If the current queue size is larger than the maximum threshold, the cell classifier 11 is sent. The first cell is discarded (step S28).

However, if the current queue size is smaller than the maximum threshold in step S27, check whether the current queue size is larger than the minimum threshold (step S29), and if the current queue size is larger than the minimum threshold, that is, the average queue size is larger than the average queue threshold. If the large and the current queue size is included between the minimum threshold and the maximum threshold, the first cell delivered from the cell classifier 11 is discarded according to the packet discard probability (step S30), and the current queue size is smaller than the minimum threshold. That is, when the average queue size is larger than the average queue threshold and the current queue size is smaller than the minimum threshold, the first cell transferred from the cell classifier 11 is stored in the corresponding VC buffers 14-1 to 14-n. (Step S26).

Here, the packet discard probability is rapidly increased by two factors. One is that the current queue size of the current VC buffers 14-1 to 14-n approaches the maximum threshold, and the probability of discarding the packets increases proportionally. One is that if the current queue size of the VC buffers 14-1 to 14-n is contained between the minimum threshold and the maximum threshold, the number of consecutively stored packets that are not discarded according to the packet discard probability increases more than a predetermined number. In this case, the discard probability rapidly increases, and the number of packets can be known by counting the AAL indication bit (AAL_indicate = 1) of the last cell constituting the packet.

On the other hand, if it is determined in step S23 that the AAL indication bit in the payload type field of the cell currently transmitted from the cell classifier 11 is found to be not the first cell of the packet, it is a cell of a damaged packet. (Step S31), if the cell transmitted currently is a cell of a damaged packet, the cell transmitted from the cell classifier 11 is discarded (step S32). It is checked whether the size is smaller than the maximum queue size (step S33).

In this case, when the current queue size is not smaller than the maximum queue size, that is, when the current queue size reaches the maximum queue size (current queue size = maximum queue size), the AAL indication bit in the payload type field of the currently delivered cell is set. After checking to see if it is the last cell of the packet (step S34), if the cell that is currently delivered is confirmed to be the last cell of the packet, that is, if the AAL indication bit of the currently delivered cell is '1', the cell classifier 11 Cell transmitted from the cell classifier 11 is stored in the corresponding VC buffers 14-1 to 14-n (step S26), and if it is confirmed that the currently delivered cell is not the last cell of the packet. The transferred cell is discarded (step S32).

When the current queue size is smaller than the maximum queue size in step S33, the cell transferred from the cell classifier 11 is stored in the corresponding VC buffers 14-1 to 14-n (step S26).

As described above, the present invention compares the average queue size with the average queue threshold for the VC buffer of the ATM switch to determine the possibility of congestion due to the overload of the ATM network, and discards the currently delivered cell according to the packet discard probability. In addition, the efficiency of the buffer can be improved to eliminate the possibility of causing total synchronization, which reduces the size of each TCP window simultaneously for multiple TCP connections.

In addition, by providing a buffer for each cell for each VC to provide VC queuing to support the FIFO service, the output of the corresponding VC buffer outputs each cell in a round-robin manner, thereby minimizing transmission loss at the packet level. This improves the injustice for bursty traffic and relatively long packets.

In addition, the embodiments according to the present invention are not limited to the above-described embodiments, and various alternatives, modifications, and changes can be made within the scope apparent to those skilled in the art.

As described above, the present invention determines the possibility of congestion on the VC buffer of the ATM switch, and discards the currently delivered cell according to the packet discard probability, thereby improving the efficiency of the buffer and inducing the overall synchronization even in a plurality of TCP connections. It is possible to eliminate the possibility, by assigning a buffer for each cell per VC to provide VC queuing to support the FIFO service, and outputs each cell in a round-robin manner at the output of the corresponding VC buffer, This can improve the unfairness of the packet.

1 is a diagram illustrating a structure for congestion control for UBR service in an ATM switch according to the present invention;

2 is a flowchart illustrating a congestion control operation for UBR service in an ATM switch according to the present invention;

FIG. 3 illustrates a VC buffer logically allocated to each VC in FIG. 1. FIG.

Explanation of symbols on the main parts of the drawings

11: cell classifier 12: input cell counter

13: switch module 14-1 to 14-n: VC buffer

15: buffer monitoring unit 16: cell service scheduler

Claims (5)

A cell classifier for classifying and transmitting each cell of the TCP packet coming into the input link by VC; An input cell counter for counting and reporting cells delivered for each VC from the cell classifier; A plurality of VC buffers for temporarily storing cells input for each VC in a corresponding buffer; A buffer monitoring unit for monitoring and reporting a state of each VC buffer; Each cell classified by VC by the cell classifier is stored or discarded in a corresponding VC buffer, and according to a cell count value reported by the input cell counter and each VC buffer status information reported by the buffer monitoring unit. A switch module for storing or discarding a currently input cell in a corresponding VC buffer; UBI service in an ATM switch comprising a cell service scheduler for outputting each cell to an output link based on a dynamic round-robin scheme in which cell slots are periodically allocated to cells stored in the VC buffers. Congestion control device for. The method of claim 1, The switch module, the congestion control device for ubiquitous service in the AT switch, characterized in that for performing the VC queuing to allocate the logical buffer for each VC to store each cell in a FIFO method. Classifying each cell constituting the TCP packet input through the input link for each VC; Checking whether the currently input cell is the first cell of the packet by examining the AAL indication bits of each cell classified by VC; Checking whether the average queue size of the VC buffer is smaller than the average queue threshold when the currently input cell is the first cell of the packet; If the average queue size is smaller than the average queue threshold and the current queue size is smaller than the maximum threshold, storing the currently input cell in a corresponding VC buffer; And discarding the currently input cell when the average queue size is smaller than the average queue threshold and the current queue size is larger than the maximum threshold. The method of claim 3, wherein Checking whether the current input cell is not the first cell of the packet as a result of examining the AAL indication bits of the cells classified by the VC; If the cell of the corrupted packet, the method for controlling the congestion of the UBI service in the ATM switch further comprising the step of discarding the currently delivered cell. The method of claim 3, wherein If the current input cell is the first cell of the packet, the method further comprises the step of discarding the first input cell when the average queue size of the VC buffer is larger than the average queue threshold and the current queue size is larger than the maximum threshold. Congestion control method for UBI service in ATM switch.