ANALYSIS OF A DISTRIBUTED ALTERNATE ROUTING STRATEGY FOR ATM NETWORKS

摘要

The paper analyzes the performance of a dynamic routing algorithm for ATM networks. The algorithm, called AR-DLCP (Alternate Routing-Distributed Least Congested Path), is based on a distributed computation, where the "best" route is chosen step-by-step by a messenger packet, which takes decisions on the basis of a cost function, composed by a local and an aggregate part. The local cost is constructed for each outgoing link, on the basis of the knowledge of the current and the maximum number of connections that the link can support, while still ensuring the required Quality of Service at the cell level. The aggregate cost is aimed at reflecting the congestion situation of a node, and is computed through an information exchange mechanism among adjacent nodes. In a fully connected network, only direct and two-hop paths are considered; in the general meshed topology case, paths are organized in a two-level hierarchy. Besides, both a static and a dynamic trunk reservation scheme are introduced to guarantee enough bandwidth to direct paths. The performance of the two alternatives is evaluated by simulation under various traffic load situations. The capacity to detect congested areas is examined, as well as the convergence of the aggregate cost computation in stationary conditions. AR-DLCP is compared with other algorithms already in the literature as DLCP, Learning Automata and RTNR (Real Time Network Routing).