Modeling and distributed simulation of complex broadband ISDN networks under overload on loosely-coupled parallel processors

Abstract

In a loosely coupled parallel processor system, a number of concurrently executable processors communicate asynchronously via explicit messages over high-speed links. This architecture is similar to broadband ISDN networks and, thus, constitutes a realistic testbed for the modeling and distributed simulation. Every switching node of the communications network is modeled on a processor of the parallel processor and the traffic flow is represented by the communications protocols. Given that the models of the switching nodes are executed in parallel, the distributed simulation is expected to execute significantly faster than the traditional uniprocessor approach. This approach makes it possible to realistically model overloads in networks, study their impact on the network performance, and investigate new schemes to limit overloads from causing uncontrollable performance degradation. An implementation of the approach on the ARMSTRONG parallel processor at Brown University is described. Performance results, based on representative example B-ISDN networks, realistic traffic models, and network characteristics indicate that the distributed approach achieves significant speedup over the traditional approach. The utilization of the traffic capacities of the links is discussed.<>