An architecture and coherency protocol for use in a large SMP computer system includes a hierarchical switch structure which allows for a number of multi-processor nodes to be coupled to the switch to operate at an optimum performance. Within each multi-processor node, a simultaneous buffering system is provided that allows all of the processors of the multi-processor node to operate at peak performance. A memory is shared among the nodes, with a portion of the memory resident at each of the multi-processor nodes. Each of the multi-processor nodes includes a number of elements for maintaining memory coherency, including a victim cache, a directory and a transaction tracking table. The victim cache allows for selective updates of victim data destined for memory stored at a remote multi-processing node, thereby improving the overall performance of memory. Memory performance is additionally improved by including, at each memory, a delayed write buffer which is used in conjunction with the directory to identify victims that are to be written into memory. An arb bus coupled to the output of the directory of each node provides a central ordering point for all messages that are transferred through the SMP. The messages comprise a number of transactions, and each transaction is assigned to a number of different virtual channels, depending upon the processing stage of the message. The use of virtual channels thus helps to maintain data coherency by providing a straightforward method for maintaining system order. Using the virtual channels and the directory structure, cache coherency problems that would previously result in deadlock may be avoided.

A receive buffer management system associates a virtual buffer pool with each node communicating with a receiver and creates an actual buffer pool for use by all nodes, with a "low-water-mark" indicating buffers are running out and a "high-water-mark" indicating enough buffers are available. Each time a buffer is taken a count is added to the virtual pool for that sending node and each time a buffer is returned to the actual pool, the counter for the sending node's virtual pool is decremented. Each virtual pool has a quota. Buffers are allocated until the number of buffers in the actual buffer pool drops below the low-water-mark. Then packets from a node above its quota will be discarded and those buffers will be immediately returned to the actual pool. Packets will be discarded for all over-quota nodes until those nodes drop below their quota or the actual pool reaches the high-water-mark. Alternatively, a sliding window acknowledgement replaces the virtual pool and counter. A receiver guarantees a transmitting node some maximum number of unacknowledged packets. A low-water-mark indicates when buffers are running out, and a maximum-locked-threshold specifies the maximum number of buffers that can be locked by the other local users. Requests above this will block. A receiver finished with a buffer returns it. When available buffers rise above the low-water-mark, acknowledges and buffer requests are enabled. Ensuing acknowledges enable transmission from waiting nodes.

A congestion control mechanism for use with a receiver in a telecommunications network using the Service Specific Connection Oriented Protocol (SSCOP). The receiver is disposed in an access network portion of the telecommunications network such that it receives messages from an aggregate portion of the network over an Access Network Interface (ANI) and from a plurality of users over User-Network Interfaces (UNIs) in a distribution network portion. Credit windows granted by the receiver to the transmitters for transmission of message frames are managed by the receiver when it experiences congestion. The congestion control method monitors buffer usage for ANI-based traffic and UNI-based traffic in the receiver by setting appropriate buffer use counters and timers. When the number of available buffers reaches certain predetermined threshold values, the receiver sends an indication to the transmitter to restrict its message transmit window, thereby throttling the message flow therefrom.

In packet switching telecommunications networks, flow control is used to obtain optimal network working point, regulating the transmitter packet sending rate. The state of overload (congestion) or underutilization of the network can be detected explicitly using signalling from network nodes, or implicitly using number of packet (window W) and round trip time (T) measurements.The Window-Time-Space Flow Control, WTFC is a method of determining the belonging part of network capacity, optimal packet sending rate and optimal window, based on the measured W,T point in the window-time space and knowledge about total network capacity W.sub.0,T.sub.0. In this way, devices with WTFC, nodes and terminals, keep optimal network working point near the on average empty queues mode of operation.With networks utilizing WTFC, nodes can signal network parameters at connection establishment only. After that, all WTFC processing is done by terminal packet transmitter. WTFC transmitter determines both optimal window and optimal sending rate, thus improving regulation stability, limiting the number of packets in the network, and decreasing the variance of transmission rate.

Method (100) executed by congestion controller (26) in each node (20, 30, 40) of a network (10) determines whether congestion is going to occur or is occurring in a particular node (20, 30, 40). If congestion is imminent or occurring, a congestion status bit is set. Each time a data packet is received and the congestion indicator bit is set, congestion controller (26) increments a congestion counter. Each time a predetermined time interval expires, congestion controller (26) decrements the congestion counter. By using the congestion counter as an index into a credit table, a credit value is determined by destination node (31) that represents how many data packets are permitted to be sent from source node (30) to destination node (31). The credit value is sent to the source node via a message.