Robust optimization of Random Early Detection

Abstract

Random Early Detection (RED) is the most widely used Adaptive Queue Management (AQM) mechanism in the internet. Although RED shows better performance than its predecessor, DropTail, its performance is highly sensitive to parameter settings. Under non-optimum parameter settings, the performance degrades and quickly approaches that of DropTail gateways. As the network conditions change dynamically and since the optimum parameter settings depend on these, the RED parameters also need to be optimized and updated dynamically. Since the interaction between RED and TCP is not well understood as analytical solutions cannot be obtained, stochastic approximation based parameter optimization is proposed as an alternative. However, simulation based approaches may yield a sub-optimal solution since for these to work, the network needs to be accurately simulated which is, however, infeasible with today's internet. In this paper, we present an optimization technique for optimizing RED parameters that makes use of direct measurements in the network. We develop a robust two-timescale simultaneous perturbation stochastic approximation algorithm with deterministic perturbation sequences for optimization of RED parameters. A proof of convergence of this algorithm is provided. Network simulations, using direct implementation of the algorithm over RED routers, are carried out to validate the proposed approach. The algorithm presented here is found to show better performance as compared to a recently proposed algorithm that adaptively tunes a RED parameter.