<title>Evaluation of ABR traffic control schemes for ATM LAN and WAN: effects of congestion indication schemes and ER switch algorithms</title>

Abstract

The ATM Forum has adopted rate-based congestion control for ABR (available bit rate) traffic. Much of the existing work evaluating ABR congestion control schemes has used some threshold value on buffer queue length to indicate congestion. On the other hand, many ER (explicit rate) algorithms calculate their 'fair-share' values based on utilization level, with the assumption that ER switches are able to measure the current utilization level of ABR traffic. If one would use the same mechanism -- measuring utilization level -- to indicate congestion, then the same switch could easily implement both binary and ER ABR control algorithms. Based on the above observations, in this paper we study the effect of using two different congestion indication methods: (1) buffer queue length (the most commonly used method); and (2) utilization level (the new method). We evaluate two binary ABR control schemes: EFCI (explicit forward congestion indication) and CI (congestion indication) using backward notification, using the two different congestion methods. We also evaluate and compare two ER algorithms: the ERICA (explicit rate indication for congestion avoidance) algorithm proposed by Jain and the CAPC-2 (congestion avoidance with proportional control - 2) algorithm proposed by Barnhart. Performance evaluation are carried out by computer simulation. We simulate two ABR switches connected by an OC-3 link, with each switch connecting five end-systems. The distance between the two switches are 20 km for LAN and 1,000 km for WAN, based on ATM forum specification. For each simulation run, we measure average queuing delay, maximum queue length, and network utilization. Traces of ACR (allowed cell rate) and buffer queue length are also examined. We found that using the new congestion method indication dramatically reduces the maximum queue length and average queuing delay, with a slight decrease in utilization. Both ER schemes show smooth buffer occupancy and attain high utilization.