This paper shows that a network supplying variable bit rate services can be prevented from becoming congested if each terminal limits the capacity of its connection in terms of its rate of increase. Variable bit rate sources are adequately assessed with two new concepts: the bit rate increase per unit time (acceleration-rate=αbit/sec2) or the bit rate increase ratio (acceleration-ratio=exp (β) ). The dimension of the acceleration-ratio coefficient βis seconds-1. The upper limits α and β are regulated to guarantee the network's QoS. The proposed concepts allow the network state to be accurately estimated and avoid congestion. The proposed method can be applied to ATM networks, Frame Relay networks, Fast Reservation Protocol systems and so on.

CEFLAR is one way of realizing ATM-ABR with no cell loss. This paper shows that the transition characteristics of CEFLAR(transition time to achieve fair share), important when addressing network fairness, strongly depend on the acceleration-ratio coefficient, not the rate decrease factor or the distance between source and congestion estimation nodes. This paper also shows that the average throughput of a transmission line in transition degrades as the rate decrease factor decreases and as the distance between the source and congestion estimation nodes increases.

Experimental demonstrations of IP over ATM with congestion avoidance flow control are described. The results confirm that this network can control the transmission rate without upper layer anomalies, IP packet loss, or TCP retransmission. Efficient and fair sharing of the available bandwidth is realized in local area networks.

A self-healing method to overcome memory address errors due to bit errors in a shared buffer memory switch is proposed. A shared buffer memory switch is one of the most useful switches for asynchronous transfer mode (ATM) networks because of its highly efficient memory usage and small hardware size. However, this switch has a basic problem. This is due to switching control mechanism in which the memory addresses are circulated between the idle address queueing buffer and the assigned address queueing buffers. Once a bit error occurs in the address, duplicate address values may appear and permanently circulate. This causes frequent cell discard and mis-routing due to cell overwriting at the same address. The proposed self-healing method can detect and discard the erroneous address and correct the address error without reducing throughput. To detect an erroneous address, a check memory is introduced. A buffer identifier in which an address is queued is memorized in the check memory. If duplicate address values appear, the memorized identifier is overwritten and changes into different one. Therefore, the erroneous address is detected by identifier mismatch when it is read out from the queueing buffer. Any queueing buffer is prohibited to store the erroneous address and that address is discarded. The correct address is restored by searching the lost address and inserting it as a correct idle address. The performance of the proposed method is simulated. It shows that the number of cells lost during the error detecting process is negligible.

Experimental demonstrations of congestion avoidance network are described. These test results show the feasibility of multi-media services in wide area networks; no congestion and fair band sharing are achieved with almost 716km optical fiber propagation delay.