Experimental Evaluation of ATM Congestion Control Mechanisms

Tt is widely believed that closed-loop per-VC flow control is needed to effectively deal with congestion for ABR traffic in ATM networks. The many ATM flow control mechanisms that have been proposed can be divided in two classes. A first class consists of roundtrip-based rate control algorithms in which a source selects a transmit rate for each VC, based on feedback from the switches along its path [2, 4, 11, 3]. Feedback can for example consist of the EFCI bit in the ATM cell header, or of an explicit rate calculated by the switches. A second class consists of hop-by-hop, credit-based flow control, in which the receiver on each link provides the sender with information on buffer status for each VC. Examples include DEC AN2 [12] and Credit Net [9, 8]. An overview of these algorithms can be found in [6, 13]. A class of rate-based algorithms is currently being standardized by the ATM Forum [14], although several companies are still pursuing a credit-based alternative called Quantum Flow Control [5]. The various flow control algorithms have been evaluated and compared extensively using simulation, for example using the ATM Forum test configurations [1, 15]. Experimental results are however limited and focus on persistent sources [16, 10]. In this paper we present an experimental comparison of several ATM flow control mechanisms using a more diverse set of benchmarks including both persisten and busty sources for a single-congestion link configurations.