WO1991015909A1 - Call admission control method and arrangement in a packet communication system - Google Patents

Abstract

A data communication system includes a data bus (1) and a number of nodes (2, 3) connected to the bus (1). Each node (2, 3) is arranged to communicate fixed length data packets via the bus (1). The data packets are assigned different priority levels and each node (2, 3) is arranged to concede access to the bus (1) to another node transmitting a higher priority level packet. Some of the nodes are modified to monitor their respective bus access rates. Each of the modified nodes (2), when its access rate falls below a respective predetermined threshold, transmits a restriction request signal onto the bus (1). The other modified nodes (2) reduce their respective bus access rates in response to the restriction request signal, thereby freeing space on the bus (1) for use by the node (2).

Description

Call admission control method and arrangement in a packet communication system

The present invention relates to a communication system suitable for use, for example, in a metropolitan area network (MAN). It is particularly, but not exclusively, concerned with a network using a Distributed Queue Dual Bus (DQDB).

Known systems such as DQDB provide for the communication of packets of fixed lengths between nodes connected to a link, in this case a bus. The packets are assembled from multiple data streams received from one or more terminals connected to the node.

Such systems generally include some mechanism for controlling the access of the nodes to the bus. The DQDB system, for example, uses two buses whose signals run in opposite directions carrying user data and network signalling information in a series of fixed length packets inserted into slot positions on each bus. A queuing discipline is used, whereby the packet at the head of the queue of data in each of a number of nodes forms, together with the corresponding packets at the heads of the queues in the other nodes, a distributed logical first come first serve queue.

While queuing disciplines of this sort may be sufficient for some types of service it is not adequate where there is a need to set on each node a different guaranteed minimum packet transfer rate onto the bus. Such a need arises, for example, with voice connections where for each node there may be a different number of active connections established. The minimum packet transfer rate will then vary according to the number of active connections.

Some attempts have been made to improve systems such as DQDB, notably through the use of counters on each node which force the node to allow a certain ratio of empty slots to pass by unused. For example, a node may be allowed to use eight empty slots and always let the ninth slot pass by. This ratio may be varied on different nodes to attempt to give each node its own unique access rate. However simulation results show that the technique is slow to converge to the desired access rate and it is difficult to fine-tune the allocations to each node.

According to a first aspect of the present invention, there is provided a communication system including a link and a number of nodes connected to the link, each node including means to communicate packets with other nodes via the link, in which at least one node includes means to measure the availability of the link capacity for the transmission of packets by the said node, comparison means to compare the measured availability with a threshold for that node and reduction means for generating a signal for reducing the utilisation availability by other nodes when the availability at the at least one node is below the threshold as determined by the comparison means.

The present invention provides a communication system with nodes capable individually of monitoring the capacity of a link such as a bus. It is therefore particularly suitable for use with critical services requiring a minimum packet transfer rate.

Preferably the said means to measure include an access rate monitor responsive to the rate at which the respective node outputs packets onto the link, and the reduction means comprises means to output a restriction request signal to enforce restriction of the output of other nodes when the rate falls below the threshold for that node. Preferably the said node is connected to a terminal adapter and the threshold is determined in accordance with the minimum packet transfer rate required by the terminal adapter.

A terminal adapter packetises the terminal data stream into fixed length packets and depacketises information received from the network, converting it to a form suitable for sending to the terminals. In the preferred aspects of the present invention, the minimum transfer rate is ensured by monitoring the access rate of the node and, when it approaches a minimum value, transmitting a control signal which causes other similarly modified nodes on the network to restrict or "throttle" their own data access rates. This results in there being more free slots on the link so that, in accordance with whatever priority or contention mechanism is adopted by the system, the node making the request is able to increase its link access rate. The system of the present invention effectively provides a higher level access control protocol for the modified nodes, the higher level protocol operating in conjunction with the basic level protocol for the system thereby ensuring complete compatibility with the basic level protocol and any unmodified nodes using that protocol. The provision of full backwards compatibility with existing systems is a major advantage, and in this respect^' the present invention contrasts with previously proposed systems, such as the Orwell system described in the applicant' s earlier application number EP-A-0, 168, 265. This is a system which while it is able to provide a guaranteed minimum bandwidth for terminals attached to the nodes, requires a dedicated control protocol. Nodes within this system have their own internal mechanism for setting their own access requirements and use available empty slots to seek additional slot allocations.

A further advantage of the present invention is that it allows efficient use of the available bandwidth on the link by services of different bit rates, including services of variable bit rates such as a variable bit rate video service. Again this contrasts with previously proposed systems such as DQDB with isochronous slots. The use of isochronous slots, that is the provision of slots of fixed, equal duration for use by different nodes, is poorly adapted to coping with a variable bit rate and in particular provides reduced efficiency through partial filling or large delays with low bit rate services such as voice telephony.

Preferably the said node includes means to receive and process a restriction request signal from another node, and the terminal adapter includes means responsive to a control - A -

signal from the said means to receive and process to switch the terminal adapter between a first mode in which the flow of data via the terminal adapter towards the network is unrestricted and a second mode in which the flow of data towards the network is restricted.

Preferably the terminal adapter includes a data buffer, when the terminal adapter is in the second mode, the data buffer being enabled to output data only in response to an output request signal from the node or after the lapsing of a predetermined delay from the initiation of the second mode.

Preferably the terminal adapter includes a register arranged to store a value d, where d is the number of packets to be transmitted to the node in response to each output request from the node, the value for d and the predetermined time being chosen to provide the required minimum packet transfer rate for the terminal.

Preferably the communication system includes nodes of first and second types, at least two of the said first type nodes including the means to measure availability, and at least one node of said second type without the capability of responding directly to the signal generated by the reduction means from another node, the or each second node being arranged to concede access to the link to first type nodes transmitting higher priority level packets.

Such second (or unmodified) type nodes are found on existing DQDB protocols and hence can be acσo odated in the present system configuration.

Preferably the communication system includes at least two sub-networks connected to a central switch for the system via respective head-end nodes, the head-end nodes being unmodified nodes and each sub-network being connected to its respective head-end node by a bridge node.

A bridge node monitors its link access rate and responds to restriction request signals. In a mixed network including unmodified head-end nodes it interfaces its respective sub-network to the system in such a way that the system as a whole remains responsive to restriction requests from individual nodes.

According to a second aspect of the invention there is provided a method of controlling a communication system including a number of nodes which can communcate via a communication link utilising packets, said method including measuring on at least one node the availability of link capacity for the transmission of packets by the said node, comparing the measured availability with a threshold for that node and generating a signal for reducing the utilisation availability by other nodes when the availability at the at least one node is below the threshold as determined by the comparison step.

According to a third aspect of the present invention there is provided a node for connection to a link in a data communication system including an output buffer arranged to hold packets for transmission onto the link, and an access rate monitor including means to determine from the packet through-put of the output buffer a parameter indicative of the link access rate available to the node.

Preferably the access rate monitor includes timer means arranged to measure the cumulative delay of the data packets in the output buffer waiting for access to the link, counter means arranged to count the number of packets output from the buffer onto the link, and processor means responsive to the timer means and the counter means and arranged to determine the said parameter from the number of packets output while the cumulative delay measured by the timer means is less than a predetermined value.

This second aspect of the present invention provides a node which has the advantage that it is able to measure the link access rate using a method which increases in speed of response as the available link capacity decreases. When there is a high link capacity available for the node then it takes longer for the cumulative delay measured by the timer means to reach the predetermined period and so determinations of the number of frames output are made only relatively infrequently. As the available link capacity decreases so the determinations are made more frequently and the responsiveness of the system increases.

According to a fourth aspect of the present invention, there is provided a method of operating a node in a communication system, including a link and a number of nodes connected to the link, including the steps of measuring the availability on the link of capacity for the transmission of packets by the node, comparing the measured availability with a threshold for the node and reducing the utilisation by other nodes if the availability is below the threshold.

A system in accordance with the present invention will now be described in detail with reference to the accompanying drawings, in which:

Figure 1 is a block diagram of a DQDB data communications system;

Figure 2 is a block diagram of a modified DQDB node;

Figure 3 is a block diagram of a terminal adapter for use in the system of Figure 1;

Figure 4 is a block diagram of a node adapter;

Figure 5 is a block diagram of a modified node combined with a terminal adapter;

Figure 6 is a block diagram of a system including sub-networks connected via bridge nodes; and

Figure 7 shows an alternative embodiment.

A DQDB data communications system includes a dual bus 1 and modified nodes 2 and unmodified nodes 3 connected to the bus 1. The unmodified nodes 3 communicate data with the bus in accordance with the DQDB protocol described in the proposed standard "Distributed Queue Dual Bus (DQDB) Metropolitan Area Network (MAN)", IEEE 802.6.

Each modified node 2 is connected to a terminal adapter 4 which is in turn connected to a terminal 5 for a service such as voice telephony requiring a guaranteed minimum data rate. Although in the present example only one terminal is connected to each adapter 5 and only one adapter 5 to each node 2, where necessary, several terminals may be connected to each adapter and or several adapters to each node. Data is communicated between the adapter 4 and the node 2 in the form of ATM (asynchronous transfer mode) packets. The existing standard for the ATM frame format includes four previously unassigned bits. It has been proposed that these bits hould be used to provide Generic Flow Control (GFC), whereby a node is able to transmit control signals to, e. g. , a terminal adapter, to modify the output from the adapter. In the system of the present embodiment the GFC bits are used to throttle the output from a given terminal adapter as described in further detail below.

As described below with respect to Figure 4, the modified node may in some circumstances be formed by the addition of a node adapter to an existing unmodified node. Also, as described below with respect to Figure 6, a combined modified node and terminal adapter may be used.

The different nodes 2, 3 contend for the bus using a contention system which, in the present example using the DQDB protocol, takes the form of a distributed first come first service queue with a place in the distributed queue assigned to each foremost packet in the internal queues of the different nodes. The modified DQDB nodes 2 are arranged to monitor the bus access rate achieved in accordance with this contention system.

Ideally, the modified node 2 monitors the rate of available empty slots passing the node. An available empty slot is one which the node can use to transmit data. Not all empty slots are available because of the internal DQDB protocol.

Again, ideally, the modified node 2 compares this available empty slot rate against the minimum requirement ' d' . When this approaches a certain level, e. g. available empty slot rate is less than 1.25d, the modified node 2 broadcasts to all other modified nodes 2 to commence throttle. Previously, according to proposed settings for the GFC the terminal adapters 4 perceive an interface state equivalent to ' unrestricted access' . Now the state of the interface is changed into ' throttled access' controlled by a reset rate.

In practice in the present embodiment the modified node 2 does not directly monitor empty slots on the DQDB bus 1. Instead it uses the departure rate of full packets onto the bus 1 to determine the available empty slot rate. This relies on the fact that when the queue of packets waiting for bus access is non-zero, the departure rate is a direct indication of the available empty slot rate.

As shown in Figure 2, two FIFOs 6 are provided within the node, one for each bus direction. The FIFOs 6 hold packets waiting for access to the bus. There is associated with each FIFO a timer 7 which is enabled whenever there is a packet in the FIFO waiting for access to the bus. The timers 7 are connected to a clock 8 provided within the node and so keep a cumulative total of the time spent by the packets in the FIFOs 6 waiting for access to the bus. There is also associated with each FIFO a counter 9 which is incremented each time a packet departs from the FIFO onto the bus 1. When the total time recorded by the timer 7 reaches a predetermined value, which in the present example is 125 microseconds, then it transmits an interrupt to the control processor 10. In response to this interrupt the processor 10 reads the value of the counter 9 into a register R and both the timer and counter are reset. The processor also includes a d register which holds the value d for the corresponding terminal.

The value of R is compared with a parameter proportional to d, which is in turn dependent on the minimum packet transfer rate needed by the terminal. If, for example, R is less than 1.25d (rounded up to the nearest integer) it implies that the terminal cannot gain access to more than 1.25d empty slots per 125 microseconds. If d slots per 125 microseconds is the minimum data rate then 1.25d represents the threshold value at which the bus access rate is considered to be too low. Accordingly the control processor generates a restriction request signal which is broadcast to the other nodes to cause them to restrict or ' throttle' the rate at which they are putting data onto the bus.

The control processor uses a packetiser 11 to output a restrict request signal in the form of an appropriately headed ATM packet which is then output onto the bus via an ATM processor 12 and one of the FIFOs 6. As described in further detail below, in response to such a request the other modified node 2 switches into a mode in which the terminal, instead of having unrestricted access to the node, transfers data only in response to a request generated by the node. The terminal adapter 4 includes its own processor 13 which receives incoming GFC bits, and in response generates an appropriate input for the upstream access control logic 14. A timer 15 provides a further input. The upstream access control logic 14 controls the flow of data through buffers 16, 17. The packet containing the restriction request signal received by the terminal adapter also carries a value indicating a reset rate in the range for the present example, from 32 per 2 milliseconds (two resets per 125 microseconds) down to 16 per 2 milliseconds (the minimum guaranteed). The reset rate is transmitted to a remote terminal adapter using a 4-bit field setting in the header of an ATM slot. The 4-bit setting (i. e. GFC field setting) is provided from the node processor 2 at the correct rate. Any such setting is written into the header of the next slot (full or empty) travelling downstream towards the terminal adapter 4. For intermediate values of R, a corresponding proportional value for the reset rate is calculated.

The following algorithm is used to translate the value of R into a reset rate: it may be noted that a reset rate of 20 per 2 milliseconds corresponds to an allocation of 1.25d.

-for d < R < 1.25d the reset rate is determined from the linear equation

-reset rate = 16R/d

-for R < d the reset rate is always 16 per 2 milliseconds -once restricted access has commenced (i. e. a node has detected a value of R < 1.25d) it will continue while R remains less than 2d and the reset rate is determined by the equation reset rate = 16R*/d, for R* < 2d where R* is the smallest value of R observed over a millisecond. A new message is sent each millisecond with the revised reset rate and originating node identity.

-if R* is greater than 2d no revised reset rate is sent.

If another modified node 2 independently broadcasts a restriction request requiring a different reset rate then the modified nodes 2 are arranged to take the most stringent of the two rate restrictions as the current setting. Throttled access continues until a time out occurs on each modified node 2 indicating that no throttled access message has been received for 3 milliseconds. This timeout is sufficiently long so that a corrupted revised reset message will not cause exit to unrestricted access provided the next revised reset message is received. Once the 3 millisecond timeout has occurred no further resets are transmitted to the terminal adapter 4 until a new message is received from one of the modified nodes or a value of R < 1.25d is detected.

The terminal adapter 4 may be arranged to operate at two different reset rates in which case the restriction request control packet carries two corresponding values for required reset rates applying to different service classes. They are distinguished by an additional class of service field. The different resets can have different minimum rates e. g. a rate of 16 per 2 milliseconds (equivalent to 1 per 125 microseconds) can be maintained for one group of services, while another reset with a minimum of 1 per 6 milliseconds could be maintained for low bit-rate services which provides a more severe throttle for the latter group (note that constant bit-rate services such as 64 kbit/s voice are not throttled since their minimum allocations are set at 64 kbit/s).

In this case the processor receives the incoming advised reset rate and calculates two or more rates at which it will provide GFC settings towards the downstream terminal adapter. For example a second type of reset could have a rate determined by the equation reset rate₂ = [(9/16) x reset rate- - 8]/3 where 32 per 2 ms > reset ratβ_j > 16 per 2 ms and

10 per 6 ms > reset rate₂ > 1 per 6 ms The appropriate GFC settings distinguishing between the two types of reset are provided at the correct rates by the node processor.

In a further modification to this procedure two types of reset can be provided to the terminal adapter to indicate the bus direction which has restricted access. In this case the node processor sends either one type of reset if only one bus has restricted access, or two types if both buses simultaneously have restricted access. This modification can be exploited by terminal adapters which employ two internal buffers which are multiplexed together to provide the output to the node as described in more detail below.

The terminal adapter terminates the GFC protocol in the downstream direction (i.e. network to terminal). It maintains a ' d' counter whose value is negotiated at call set-up time. Call control will ensure that no call is accepted if the sum of all ' d' s exceeds a predefined level of the bus capacity (say 90%). This implies that when throttled to maximum the terminals will each receive their guaranteed share.

Call control responds to a call request message with information which enables the packetiser in the terminal adapter to associate a correctly labelled header with the paσketised information of the new call. This response also provides the terminal adapter with the go-ahead to change the allocation d, either via an explicit field within the message from call control or else through information stored in the terminal adapter based on the known type of terminals connected to it.

For the case of terminal adapters employing two buffers depending on the bus direction, the packet header information from call control will include information to indicate which of the two buffers should be used.

The terminal adapter will signal to the processor in the modified node any change in its d-allocation as agreed with call control. The modified node maintains a record of the current d-value of any terminal adapter connected to it so that it can load the appropriate value of 1.25d in its register.According to one setting of the 4 GFC bits in the downstream direction the terminal adapters will pass packets upstream in an unrestricted way.

On receipt of an ' immediate return to floor' setting of the GFC the terminals will halt transmission awaiting the first reset (another setting of the GFC). On receipt of the first reset the terminal adapters will transmit ' d' packets and halt again awaiting a further reset.

For the case of a terminal adapter employing two buffers, one for each bus direction, the change from unrestricted access to restricted access depends on the bus direction setting associated with the ' immediate return to floor' message and the associated resets. Separate controls are maintained in the terminal adapter dealing with each direction.

Upon receipt of each reset a timer is started (see Figure 3). If no further reset is achieved after 125 microseconds the timer will generate an ' auto-reset' which provides the node with a further allocation. If after a further 125 microseconds still no reset has been received via the GFC field, the terminal adapter will revert to unrestricted access.

Another modification to the system is shown in figure 4. Equipment to provide guaranteed bandwidth control may be remotely connected to some types of DQDB nodes which satisfy the following conditions: -the DQDB node has an ATM terminal interface, i. e. information is conveyed from the terminal to the node already packetised in fixed length ATM packets, -the DQDB node operates some of the GFC bits on that interface to throttle terminals under conditions of overload on either of the buses. This could be via the simple use of the ' immediate return to floor' setting which stops/reduces the transmission of packets for a known period.

In this case a node adapter may be connected to the DQDB node on a terminal interface. By using the same method of monitoring the allocation with a timer and a counter the node adapter can signal to other node adapter or modified nodes when they should enter restricted access. In this case entry into restricted access is not specific to a particular bus direction as it can be with the modified node as this information is not necessarily available from the GFC settings from the DQDB node.

For the case of the combined node/terminal adapter (see Figure 5) the need for an auto-reset and associated timer is removed since reset-controlled access from the output FIFOs is handled directly by the node processor. For this purpose an output enable signal and packet sent indication are connected to the node processor for each bus direction. The node processor permits only d packets to be sent within a period equivalent to the reset interval (i. e. the time between successive resets) when the state of restricted access has been entered. There is no requirement to set or use any GFC bits since no remote resets are required. In both the system of Figure 5 and Figure 4, the control processor 10' directly controls the output from the node by the transmission of enable signals to the output buffer 6' . A further aspect to be considered is how modified DQDB nodes can be added to a public MAN arrangement such as that shown in Figure 6. The distinguishing features of this arrangement are: -each customer has a DQDB bus connection to a central switch 18 known as the MAN switching system. This interface is known as the Subscriber Network Interface or ' SNI' . The customer can connect one or several DQDB nodes on his premises to the same SNI.

-at the network-end of each SNI, where it terminates on the central switch, is a head-end DQDB node 19. -in any arrangement where several customers are connected to the same MAN, the head-end node 19 would not normally be owned by a customer wishing to upgrade his network for multi-services. These nodes would therefore remain unmodified whenever such a customer adds modified nodes to his SNI.

There is a problem to be considered when there is an unmodified node at the head end through which all public services are sent, e. g. from customers B and C to customer A. Since this unmodified node is a source of higher priority traffic on the SNI (in the direction network to customer) it would compete for bandwidth with a group of modified nodes attached to the SNI. However it contains no throttle control unlike the modified nodes.

To overcome this problem it is necessary to configure the subnetwork as two DQDB systems connected by a bridge node 20. In the bridge node 20 there are additional modified node functions. Further modified nodes are only added to the customer-owned DQDB system.

With this arrangement the unmodified head-end node 19 at the central switch does not compete with any modified node in the customer' s premises. This can easily be seen by considering the load on each bus direction of the SNI in turn. In the network-to-customer direction only the head-end node loads the bus. In the customer-to-network direction only the modified bridge node loads the bus. Hence there is no competition for bandwidth on the SNI.

Within the customer-owned DQDB bus system, on the other hand, the bridge node 20 becomes the head-end node, but it is of the modified variety. Hence throttle controls can be applied to this head end and therefore it can operate successfully with other modified nodes added to this subnetwork.

The functions of the bridge node 20 are:

- to extract all packets from network to customer and, according to the header label, transmit the packets to an attached terminal adapter or, without modifying the packet header, organise the packets into a low priority queue and higher priority queues awaiting access to the customer-owned DQDB bus.

- to monitor the access rate of the high priority queue on to the customer-owned DQDB bus and to throttle other modified nodes attached to this bus as necessary according to the principles already described.

- to extract all packets from the customer-owned DQDB bus and, according to the header label, transmit the packets to an attached terminal adapter, delete packets which are destined for other terminals within the customer premises and organise the remainder into low priority and high priority queues awaiting access to the SNI (with no modification to the packet headers).

- optionally the bridge node may throttle all modified nodes (including itself) if the high priority load from customer to network approaches the capacity of the SNI. To do this the bridge node must monitor the high priority load using a separate counter and timer for this purpose. The timer would be set to interrupt the node processor after, say 6 milliseconds, and is clocked whether the FIFO is full or empty (unlike the timer used for bandwidth into the FIFO. After 6 milliseconds the counter is compared against a d value used for the SNI (e. g. d could be set to 80% of the total SNI packets within a 6 millisecond interval rounded down to the nearest integer). For measured loads in the range: d < load < 1.25d the bridge node may optionally send a reset rate to other modified nodes using the formula: reset rate = 16 x load/d and would exit from restricted access to unrestricted access whenever the measured load returns below d.

- optionally the bridge node may throttle all other bridge nodes if the measured network-to-customer load attempting to access the customer-owned DQDB bus exceeded say 80% of its capacity. This requires in the bridge node a separate timer and counter to measure the load say every 6 milliseconds with operations as described above for entry and exit from the restricted access condition.

Although the system has generally been described in relation to a DQDB configuration with modified nodes, the arrangement could be applicable to other types of bus arrangements.

Thus a dual bus arrangement, for example, using Orwell type nodes typically formed from multiplexers of they type disclosed in European Patent Publication 0168265A. but modified to operate in the manner required in present invention could be used.

With this in mind, the multiplexers, in addition to handling the queues at any one node and determining their d values for the various priority queues, will additionally include the mechanism for sensing the bandwidth availability on the bus at any time and to request reduction of utilization of other nodes if availability is unduly limited.

Thus the Figure 1 arrangement described above could include Orwell type multiplexers forming each node and all/some of these nodes being configured to include the ability to measure and request reduced utilization from other similar nodes.

Such an arrangement is now described below with regard to Figure 7 which includes a number of nodes 100. As shown three nodes 100 are within the Customer premises network connected by means of dual bus 103 at the Broadband ISDN user network interface (BISDN-UNI) to the public BISDN network with a further node 100 being shown within BISDN switch 101.

Each node 100 independently monitors the utilisation level of the bus (in each direction) and when this approaches a critical level it will set the generic flow control (GFC) field of the BISDN packet headers (known as ATM cells) so as to restrict the access rate of all nodes.

A key feature of the GFC mechanism for restricted access is that it provides a certain minimum guaranteed bandwidth available to each node. This relies on the inclusion of a counter in each node, whose value is the number of packets which the node can send between successive GFC signals known as ' resets' . Clearly the value in the counter can differ on each node, according to the needs of the service. Furthermore the protocol ensures that the reset rate is never less than one per 125 microsecs, which therefore represents the minimum guaranteed bandwidth for the node.

To perform a bandwidth monitoring function, each node estimates how much bus bandwidth is currently available to it as follows. For each bus direction, whenever a packet is waiting for bus access timing pulses are fed to a timer T. No timing pulses are counted when there is no packet waiting. For each bus direction a counter C is enabled to count the number of cells departing onto the bus. When T equals 125 microsecs C is read into a register R and the timers and counters are reset.

The value in register R is an estimate of the bus bandwidth available to the node (expresesed as the number of cells that can be output in 125 microsecs).

Nodes are configured to be a particular ' access state' dependent on bandwidth utilisation conditions as now described. The minimum guaranteed access rate for a node is stored as the value d cells/125 microsecs. If the independently measured value R on each node remains greater than threshold M (eg M = 1.25d, ie 25% above minimum) no throttling action is taken. In this case all nodes are in a state of ' unrestricted access' . However if, for any node, R is less than M on one of the two busses, it implies bandwidth is near the minimum acceptable and this node sets the GFC field to change the state of all nodes to ' restricted access' . This state applies to a particular bus and all nodes are limited to an access rate of d packets per ' reset' signal. The node which originated the restricted access signal becomes the ' master node' determining the reset rate. Each reset signal is placed on the opposite bus by the master node and is then re-issued by the head end node on the correct bus. The reset rate may vary from N per millisec (eg 16 resets per millisec) down to 8 per millisec (the minimum guaranteed).

While in the restricted access state the master node continues to monitor the bus access rate and maintains the restricted access state while: reset rate < N per millisec

If the reset rate is greater than N, the master node sets GFC field to return all nodes to ' unrestricted access' . Once again the signal is passed down the opposite bus and is reflected at the head end.

A node may issue a ' restricted access' signal even when it is already in the restricted access state. This is when it wishes to take over as master because it observes that its measured value R has fallen below M. It issues a ' restricted access' signal and this causes the current master to relinquish control as explained further below. The new master is immediately free to use its own slower reset rate to provide more bandwidth for it to use.

It may happen that two or more nodes measure a low value of R and attempt to become master. This is resolved as follows. When a node writes the ' restricted access' setting into the GFC field it begins to monitor both busses and remains master until either:

- it exits to the unrestricted access state (see above)

- it reads that another node has set the ' restricted access' state for the appropriate bus direction. The earliest warning it could receive is by looking at the opposite bus for a ' restricted access' signal which is destined to be reflected by the head end (signals which are to be refelcted are coded differently in the GFC field from those which are not reflected).

While current master (of particular bus direction) a node reads the opposite bus and overwrites any reset signal from other nodes if the GFC setting indicates that such reset signals are to be reflected at the head end.

The issue of resets and response to resets is determined by a protocol now described.

Following the issue of a ' restricted access' signal the master node immediately issues a ' reset' signal and simultaneously resets its own counter. For each cell which it transmits on to the bus for which it is master it decrements the counter by 1. Then if either:

- the counter reaches zero, or

- no cell is waiting for transmission the master issues a further reset signal and immediately resets its counter. When these reset signals return via the head-end they are ignored by the master node (but not by other nodes).

Following the receipt of a ' restricted access' signal by other nodes they cease transmission awaiting the first reset signal. When this arrives each node resets its counter by one each time. When the counter reaches zero the node ceases transmission until either:

- a timeout occurs indicating that 125 microsecs have elapsed since the last reset. In this case the node issues an ' auto-reset' and resets its own counter and sets its own state to ' auto-resetting' .

- a further reset signal arrives and the node state is not ' auto-resetting' .

If a reset arrives and the node state is ' auto-resetting' the effect of this signal is to cancel the ' auto-resetting' state. The reset timer continues to measure the time elapsed since the last auto-reset. If no further reset arrives whilst the node state is auto-resetting then the occurrences of a further time out of 125 microseconds since the last autoreset causes the node to enter the unrestricted access state.

The possibility of auto-resetting ensures that each node is able to transmit at least at the minimum guaranteed rate.

GFC field settings are dealt with as now described.

Each GFC signal is transmitted via the head end where it is modified to show the particular bus direction to which it applies. Any signal is written into the header of a full or empty slot.

In one mode of operation, nodes can be allowed access to the bus in an unrestricted way. This is signified by a continuous series of ' null' GFC settings.

However, on receipt of a ' change access state restricted' setting of the GFC the nodes will halt transmission awaiting the first reset.

On receipt of a ' change access state-unrestricted' setting the nodes return to unrestricted access.

These four signals (reset, null, change access restricted, change access unrestricted) are coded as follows:

- On the bus direction to which they apply they are coded using the lower 2 bits.

- On the opposite bus the reset and change access state signals (3 signals) are coded by overwriting the other 2 bits of any GFC field where the lower 2 bits are already set to read ' null' and the upper two bits are also set to ' null' .

Optionally the protocol can take account of operations on a service class basis as follows. Change access state signal and reset signals are supplemented with information about the service class in the upper two bits, ie.

- 00 = ' lower class'

- 01 = ' lowest two classes'

- 10 = ' lowest three classes'

- 11 = ' all classes' Service class information only appears for these 3 signals (reset and change access state signals). For the null signal, the upper two bits contain information for the opposite bus.

The head end node operates more intelligently by trying to decide which service class to append to the signal (no service class information is appended to signals until they reach the head end). For example on entry from the unrestricted access state, the head end could append ' lowest class' only. It could then successively change the class parameter until the reset rate is at least greater than say 8M/d per millisec indicating that the master node is now above the threshold ' M' discussed above. If a second ' restricted access' signal is received while in the restricted access state it could immediately change the class parameter to restrict more classes.

Alternatively the response of the head end node to more than one ' restricted access' signal can be exploited as follows. The master node can shift the class parameter to the desired mode by sending a sequence of x ' restricted access' signals (x is not more than 4) prior to sending the first reset. A copy of the current class mode is kept in each node. It would need to be kept in synchronisation with other nodes by an occasional broadcast signal from the head end. However, the nodes are informed of the current class of service status at the head end wherever a restricted access signal or unrestricted access signal is dispatched from the head end. In this arrangement it would be of benefit if the bandwidth monitoring function is replicated in each node on the basis of one per service queue.

The system just described allows each node to independently monitor its available bus bandwidth and to do no more than this until the bandwidth drops below a measured level. Until such a level is reached there is no restriction on bus load which can be important on long busses, for example. When restriction is deemed necessary, each node is guaranteed at least a minimum bandwidth.

Claims

1. A communication system including a link and a number of nodes connected to the link, each node including means to communicate packets with other nodes via the link, in which at least one node includes means to measure the availability of the link capacity for the transmission of packets by the said node, comparison means to compare the measured availability with a threshold for that node and reduction means for generating a signal for reducing the utilisation availability by other nodes when the availability at the at least one node is below the threshold as determined by the comparison means.

2. A system according to claim 1, in which the said means to measure include an access rate monitor responsive to the rate at which the respective node outputs packets onto the link, and the reduction means comprises means to output a restriction request signal to enforce restriction of the output of other nodes when the rate falls below the threshold for that node.

3. A system according to claim 2, in which the said node is connected to a terminal adapter and the threshold is determined in accordance with the minimum packet transfer rate required by the terminal adapter.

4. A system according to claim 3, in which the said node includes means to receive and process a restriction request signal from another node, and the terminal adapter includes means responsive to a control signal from the said means to receive and process to switch the terminal adapter between a first mode in which the flow of data via the terminal adapter towards the network is unrestricted and a second mode in which the flow of data towards the network is restricted.

5. A system according to claim 4, in which the terminal adapter includes a data buffer, when the terminal adapter is in the second mode, the data buffer being enabled to output data only in response to an output request signal from the node or after the lapsing of a predetermined delay from the initiation of the second mode.

6. A system according to claim 5, in which the terminal adapter includes a register arranged to store a value d, where d is the number of packets to be transmitted to the node in response to each output request from the node, the value for d and the predetermined time being chosen to provide the required minimum packet transfer rate for the terminal.

7. A system according to any one of the preceding claims, including nodes of first and second type, at least two of the said first type nodes including the means to measure availability, and at least one node of said second type node without the capability of responding directly to the signal generated by the reduction means from another node, the or each second type node being arranged to concede access to the link to first type nodes transmitting higher priority level packets.

8. A system according to any one of the preceding claims, including at least two sub-networks connected to a central switch for the system via respective head-end nodes, the head-end nodes being second type nodes and each sub-network being connected to its respective head-end node by a bridge node.

9. A system as claimed in claim 1 or 2 including means within each node for assuming a control status when the measured bandwidth is below the threshold, means for generating a signal indicative of said contol status, and means for broadcasting said control status signal and the restriction signal.

10. A system as claimed in claim 9 including means at the node for detecting a restriction signal from another node to revert to non-control status if the receiving node was assuming a control status.

11. A system as claimed in claim 9 or 10 including means at the node for generating resets whilst the node is in the control status, means for comparing the reset rate with a known threshold and means for generating a signal indicative of unrestricted access when the reset rate is above the threshold.

12. A system as claimed in claim 11 including means at the node for responding to resets when in the non-control status so as to reset the allocation for that node.

13. A system as claimed in any one of claims 9 to 12 including means at the node for storing the class of service for which restriction is to apply, means for generating a sequence of restriction signals and remote means on the link for responding to the sequence of restriction signals to alter the class of service for which restriction is to apply.

14. A system as claimed in claim 13 wherein the restriction signal sent to the remote means for dispatch therefrom with the appropriate class parameter is sent by a node by means of a setting in a fixed portion of the packet field.

15. A system as claimed in claim 14 wherein the packet field portion comprises 4 bits and the class parameter uses two bits with a null setting being sent to the remaining two bits, the number of restriction signals received by the remote means determining the class of service setting.

16. A system as claimed in claim 15 wherein the remote means includes means for modifying the fixed portion of the packet field to indicate the class of service using two bits, and using the other two bits for any restriction signal.

17. A system as claimed in claim 16 wherein the remote means is configured to modify the packet field to indicate class of service on receipt of a number of unrestricted access signals.

18. A method of controlling a communication system including a number of nodes which can communicate via a communication link utilising packets, said method including measuring on at least one node the availability of link capacity for the transmission of packets by the said node, comparing the measured availability with a threshold for that node and generating a signal for reducing the utilisation availability by other nodes when the availability at the at least one node is below the threshold as determined by the comparison step.

19. A node for connection to a link in a data communication system including an output buffer arranged to hold packets for transmission onto the link, and an access rate monitor including means to determine from the packet through-put of the output buffer a parameter indicative of the link access rate available to the node.

20. A node according to claim 19, in which the access rate monitor includes timer means arranged to measure the cumulative delay of the data packets in the output buffer waiting for access to the link, counter means arranged to count the number of packets output from the buffer onto the link, and processor means responsive to the timer means and the counter means and arranged to determine the said parameter from the number of packets output while the cumulative delay measured by the timer means is less than a predetermined value.

21. A data communication system according to any one of claims 1 to 8, including a node according to claim 19 or 20.

22. A method of operating a node in a communication system including a link and a number of nodes connected to the link, including the steps of measuring the availability on the link of capacity for the transmission of packets by the node, comparing the measured availability with a threshold for that node and reducing the utilisation by other nodes if the availability is below the threshold.

23. A method according to claim 22, in which the availability of capacity for the node is determined from the rate at which the node outputs packets onto the link, the method further comprising outputting a restriction request signal to restrict the output of other nodes when the rate falls below the threshold for that node.

24. A method according to claim 23, in which the threshold is determined in accordance with the minimum packet transfer rate required by a terminal adapter connected to the node.

25. A method according to claim 24, further comprising receiving a restriction request signal from another node, and generating in response to the restriction request signal a control signal to switch the terminal adapter between a first mode in which the flow of data via the terminal adapter towards the network is unrestricted and a second mode in which the flow of data towards the network is restricted.

26. A method according to claim 25, in which the terminal adapter includes a data buffer, after the terminal adapter is switched to the second mode the data buffer being enabled to output data only in response to an output request signal from the node or after the lapsing of a predetermined delay from the initiation of the second mode.

27. A method according to claim 26, in which in response to each output request the terminal transmits a predetermined number d of packets to the node, the value of d and the predetermined time being chosen to provide the required minimum packet transfer rate for the terminal.

28. A communication system substantially as described with respect to the accompanying drawings.

29. A method of controlling a communication system as claimed in claim 18 and substantially as described therein.

30. A method of operating a node in a communication system substantially as described with respect to the accompanying drawings.