AU710270B2 - A method for controlling congestion in a telecommunications network - Google Patents

Description

-2- A METHOD FOR CONTROLLING CONGESTION IN A TELECOMMUNICATIONS

NETWORK

The present invention relates to controlling congestion in a packet switched telecommunications network.

Congestion control needs to be implemented in packet switched telecommunications networks to prevent congestion collapse of a node in the network and to allow sources reasonable or fair access to network links to transmit data traffic.

°°eeoc .:.ooi An example of a packet switch telecommunication network is an asynchronous S 15 digital communications network known as the Asynchronous Transfer Mode (ATM) network. The ATM network has been developed by the international telecommunications standards authority CCITT, now known as ITU-T, the ATM Forum, research organisations and telecommunications manufacturers and providers, and is defined by a set of recommendations established by CCITT. The ATM network is a connection ooo° S 20 orientated packet switching network which carries information in fixed length packets called cells. The cells comprise 53 octets, 48 octets of source data and 5 octets comprising a header which includes signalling and addressing information. The switches of the network form network nodes and use physical and/or logical links between one another for the transmission and switching of cells. A number of node to node links and connections may be required to form a connection or path between a source terminal and destination terminal. If a logical path between the source and destination is used, then one or more "virtual paths" across the network may be required to form the connection between the two, each virtual path providing a logical link. The links, physical or logical, each carry a stream of cells multiplexed from a number of connections or sources. Multiplexing is performed asynchronously with cells being queued and interleaved onto the links without reference to any fixed multiplexing structure, such as that used in time division multiplexing.

950725,p:\oper\dbWPM7058.94,2 -3- ATM cells may be transmitted from a source in groups referred to as packets which may each comprise about 100 cells, the number varying depending on the transmission protocol being used. Transmission protocols are available for use by sources which recover corrupted or lost packets, and this is normally achieved by retransmission. If an error is found in one cell, for example, then the entire packet, including the cells successfully received, is retransmitted. Most of the protocols perform retransmission by retransmitting a number of packets at once in a retransmission window, which can lead to a situation where a link of a node is transmitting only retransmitted data. Retransmitted data is referred to as "badput", as opposed to original data transmission which is referred to as "goodput". An event known as congestion collapse occurs when a link of a node is transmitting badput without any goodput, i.e.

all of the available bandwidth of the link is consumed for retransmission. When this occurs the goodput is overwhelmed by the quantity of badput and original data traffic $oooo packets are simply discarded as they cannot get access to the link.

Reducing the size of the retransmission window provides assistance in alleviating congestion, and an adaptive window flow control protocol is available to reduce window size. The adaptive flow protocol may begin by retransmitting packets in groups of ten if an error is found in one, and when the source finds that cells are being discarded or ooe• lost at a link, the protocol causes this source to reduce its retransmission window from ten to nine packets. This reduction continues for each cell lost until the retransmission window has a size of one packet. Whilst the adaptive protocol is of assistance, it requires sources to implement the protocol and abide by traffic allocation limits. To encourage use of the adaptive protocol and fair use of network capacity, sources which exceed traffic requirements should be penalised. Penalties can be achieved by selectively discarding cells at links, which also directly relieves and controls congestion.

One method of discarding cells and controlling congestion is known as fairness queuing which involves implementing a queue for each source using a link at the output ports of a fast packet switch. The queues are served consecutively and, cells are automatically discarded when a source's respective queue is full, which would occur during excessive use by the source of the link. Fairness queuing, however, is expensive to 950725,p\opcr\dbWPM7058.94,3 -4implement in high speed packet switched networks as a separate queue needs to exist for each user of each network link. The requirement for a separate queue is particularly onerous in ATM networks where the queues have to be implemented in hardware.

Another method known as fairness discarding involves maintaining a count for every source using a link of the number of cells which a source has queued for transmission on that link. By monitoring the counts, the fairness discarding method is able to avoid congestion by selectively discarding the cells of those sources which are excessively using the available bandwidth. If the output ports each include a queue of 1000 cells then arguably the switch would need to maintain 1000 counters for each queue, to account for each available location in a queue being taken up by a different source. Therefore this solution, although equitable, can prove expensive to implement.

••oo• In accordance with the present invention there is provided a method for 15 controlling congestion in packet switched telecommunications networks, comprising discarding cells of a connection for a predetermined period when a discard state for a buffer of a link is detected.

Advantageously said predetermined period is selected to ensure the number of cells discarded is at least the number of cells in a retransmission window for said connection.

Preferably the predetermined period is at least twice the propagation delay between the source and destination terminals of said connection.

Preferably said method includes discarding cells of another connection if said discard state remains.

Preferably said method includes detecting said discard state when the number of cells in said buffer reaches a predetermined threshold.

A preferred embodiment of the present invention is hereinafter described, by way 950725,p:\oper\dbwPM7058.94,4 of example only, with reference to the accompanying drawing, wherein: Figure 1 is a schematic diagram of an output port of a fast packet switch using a preferred embodiment of a statistical discarding protocol.

Fast packet switches for telecommunications networks may be implemented using a number of different configurations. The switches may be controlled primarily by software, such as described in "Stochastic Fairness Queuing", P.E. McKenney, Internetworking: Research and Experience, Vol. 2, Pages 113 to 131, 1991, or as in an ATM network which meets B-ISDN standards, the switches may be implemented primarily in hardware, such as described in "The Application of Multistage Interconnection Networks to Fast Packet Switching", R. Palmer, C. O'Neill and E. Tirtaatmadja, Journal of Electrical and Electronics Engineering, Australia, Vol. 8, No. 2, Pages 119 :eooo° to 129, June 1988. The switches all include processing and data storage capabilities egooo which can be adjusted to alter the packet switching and transmission protocols executed by the switches.

A statistical discarding protocol, which can be used in the fast packet switches of ATM networks to control congestion, is implemented for each output port 2 of a fast packet switch, as shown in Figure 1. A cell discarder 4 is provided before a buffer 6 S 20 of the output port 2, with a threshold detector 8 being used to detect the state of the buffer 6 and provide control signals to the discarder 4. The buffer 6 acts as a "first in :first out" (FIFO) queue for cells to be transmitted out onto the link 10 of the port 2.

Cells 12 sent by the switch to the buffer 6 are received by the discarder 4 first and the discarder 4 is able to access the addressing information in the cells 12 to determine the connection or source to which a cell 12 corresponds. The detector 8 is configured to monitor the buffer 6 for a discard state, such as when the buffer 6 includes a predetermined number of cells 12. On detecting the state the detector 8 instructs the discarder 4 to begin discarding cells. The state may, for example, occur when the buffer 6 is half full.

On receiving the instruction to discard cells, the discarder 4 discards the next cell 12 received and using the address information contained in the cell determines the 950725,p:\oper\dbPM7058.94,5 -6connection or source associated with the discarded cell. The discarder 4 then continues to discard cells of that connection when received for a period of twice the propagation delay for that connection between the source terminal and the destination terminal of the connection. Discarding cells of the connection for this period ensures that instead of discarding one cell for a connection and then having the cell's entire packet retransmitted as part of a retransmission window, due to the loss of that one cell, all of the cells in the retransmission window will be discarded. Therefore the retransmission window will be retransmitting all of the cells which have been discarded for that connection. This prevents the network resources being used, especially under congestion conditions, to transmit cells which will be retransmitted anyway.

Discarding cells of a connection for the twice propagation delay period ensures the source of the connection realises that congestion is relatively severe due to the loss of a large number of cells, and this causes the source to reduce its transmission rate, S 15 thereby relieving congestion. The discarding period ensures that cells of an entire retransmission window are discarded and not one cell per window.

If, after discarding cells of a connection for twice the propagation delay, the buffer 6 is still in a state which the detector 8 recognises as a discard state, then the discarder 4 is instructed by the detector 8 to discard cells for another connection for twice the propagation delay for that connection. This step is performed iteratively until the buffer 6 is detected by the detector 8 as no longer being in a discard state.

Randomly selecting connections for the discarding of cells provides a fair method of relieving congestion, as over time, statistically an equal number of connections will be targeted if they are each adopting a similar fair transmission protocol, such as the adaptive window flow control protocol. The discarding method will, advantageously, target those connections who adopt other protocols and exceed fair transmission rates.

The statistical discarding protocol provides a simple and fair method for relieving congestion, which only requires minor software and/or hardware adjustments to a fast packet switch. There is no need for counters or queues for every source or connection 950725,p:\opr\dbWPM7058.94,6 -7using an output port of a switch.

Alternatively, rather than relying on the statistical probability of discarding packets of excessively transmitting connections, the detector 8 can be configured so as to monitor use of the buffer and select those connections which should be targeted for discarding. The detector 8 can then instruct the discarder 4 to discard cells of the specific connections for twice the propagation delay, when the discard state is detected.

Examples of connections or cells which could be selected are: 1. The connection with the source terminal closest to the switch.

2. The connection having the smallest source to destination propagation delay.

3. A connection having a low pre-assigned priority level.

oe 4. ATM cells having a cell loss priority (CLP) bit set, indicating they can be discarded during congestion.

S. Throughout this specification and the claims which follow, unless the context requires otherwise, the word "comprise", or variations such as "comprises" or "comprising", will be understood to imply the inclusion of a stated integer or group of integers but not the exclusion of any other integer or group of integers.

*o 950725,p:\oper\dbPM7058.94,7 P:\OPER\DBW\27173-95. 196- 15/7/99 -8- THE CLAIMS DEFINING THE INVENTION ARE AS FOLLOWS: 1. A method for controlling congestion in packet switched telecommunications networks, comprising discarding cells of a connection for a predetermined period when a discard state for a buffer of a link is detected.

2. A method as claimed in claim 1, including detecting said discard state when the number of cells in said buffer reaches a predetermined threshold.

3. A method as claimed in claim 1 or 2, wherein the predetermined period is at least S twice the propagation delay between the source and destination terminals of said connection.

oo.4. A method as claimed in claim 1, 2 or 3, said predetermined period is selected to ensure the number of cells discarded is at least the number of cells in a retransmission window Ii 15 for said connection.

A method as claimed in any one of the preceding claims, including discarding cells of another connection if said discard state remains after said predetermined period.

20 6. A method as claimed in any one of the preceding claims, wherein said connection is randomly selected 7. A method as claimed in any one of the preceding claims, wherein the discarded cells were destined for said buffer.

8. A method as claimed in any one of the preceding claims, wherein said buffer is for an output port of a telecommunications switch.

9. A method as claimed in claim 8, wherein said switch and said link are of an ATM network.