Date:Thursday, November 18, 1999
Time: 15:00
Place: Aula 214

Speaker: CongDuc Pham, Lab. RESAM, Universita' Lyon 1, Claude Bernard, France.
Visiting DISI in the framework of a GALILEO Italian-French research project.
Title: Parallel and Distributed Simulation of Communication Networks: towards cluster-based solutions

Abstract.

As the demand for broadband communication is becoming larger and larger it is very urging to offer high-speed backbones to provide sufficient bandwidth for a various number of multimedia applications such as video-conferencing and TV distribution. Although simulations can be carried for small or specific parts of the network (e.g. switch element for switch architecture design), the challenge is now to simulate it in one piece. In an environment where both speed and bandwidth are increased, problems like congestion avoidance, traffic control and optimal routing algorithms should be addressed first to ensure the best behavior of the network.

Simulation is nowadays widely used for system prediction and performance analysis tasks. However as the demand for larger and more complex models increases, sequential simulations, i.e. those performed on a single processor, become very time-consuming and appear less attractive than before. In order to reduce the simulation time, and to increase the system size as well, parallel simulation techniques have been developed during the last 2 decades to distribute a simulation over a set of processors. Radically different from simple replications, parallel simulation consists in partitioning a model into a set of small logical processes (LPs). These LPs are executed on different processors and interact each others by exchanging timestamped messages. The main concern for parallel simulation is to maintain the correct execution of events, i.e. in timestamp order, with a synchronization protocol. =

Parallel simulation of small grain applications have been traditionally performed on powerful parallel machines because the communication cost between the processors must be kept small in order to obtain significant speedups. The availability of high-performance clusters (HPC), at an affordable price, and adequate communication software is a great opportunity for the parallel simulation community to bring these techniques to a larger audience.

This seminar will present an overview of parallel simulation techniques and the CSAM tool for ATM network models. Experiments on a high-performance cluster will be presented.

For any additional information please contact Giovanni Chiola