A Robust Heuristic for Minimizing Cell Loss in ATM Networks
MASCOTS '95 Proceedings of the 3rd International Workshop on Modeling, Analysis, and Simulation of Computer and Telecommunication Systems
| Hi-index | 0.00 |
This paper attempts to answer the following questions related to buffer management schemes for high speed integrated services networks: ($i$) given the pattern of cell arrivals from different classes of traffic, can buffer control significantly influence the effect of cell loss ($ii$) given a performance measure in terms of cell loss, how are some or all of the characteristics of an optimal buffer control policy described and ($iii$) would an end-to-end buffer control policy (that is, a policy that is exercised over the entire network) be significantly different from one meant for controlling a single node. The basic approach to answering these questions is to impute a cost of losing cells which may vary between classes of applications and to minimize this cost over the finite or infinite time horizons. We provide a complete characterization of optimal buffer control for a single node for minimizing linear, concave and convex cost functions of cell loss. For the end-to-end problem, we show for the simple case when there is no cross traffic in the network, the intuitive approach of controlling only the first node in the series is sufficient. For the general multi-node problem, we present insights into the nature of optimal policies. Then, using a dynamic programming formulation for a two-node network, we construct a simple heuristic that significantly outperforms a control based on static rules.