High performance scalable computing performance modelling using ptolemy

The High Performance Scalable Computing (HPSC) architecture supports distributed implementations of applications requiring high performance, both in terms of processor cycles and data bandwidth. This architecture uses clusters of multi-processor nodes interconnected by a network using the MYRINET protocol. This paper describes the simulation capability developed to model and predict the performance of the HPSC architecture. This capability was created by developing various models and extensions to Ptolemy, a software modelling environment developed at the University of California, Berkeley.