Using time warp for computer network simulations on transputers

Using Time Warp for Computer Network Simulations on Transputers Fannie Tallieu, Frank Verboven Brussels Free University W B Pleinlaan 2 B1050 Brussels Belgium email: fdtallie@info.vub.ac.be In thispaper t . 2 implementation 0 i Time Warp simulation frame is examined on a multitransputer machine. As computer networks can be considered as an integratedpart of machine architecture, reliable evaluation tools for the development and study of protocols are a necessity. Simulation constitutes the only possible tool to obtain detailed information in this domain. The complexity of computer networks makes their simulation computational complex and also very slow. That is why it is interesting to consider distributed simulation approaches. A major goal of this work is to determine whether the optimistic Time Warp algorithm is advantageous for the simulation of computer networks. The results presented indicate that the eficiency of the Time Warp technique seems to depend heavily on the characteristics of the simulation model. Introduction. Simulations of large and complex systems are among the most expensive of all computational tasks. To speed up simulations, the use of parallelism appears to be a promising approach. It is possible toexploit their potential concurrency, if the simulation programs are structured in a way that they exhibit this potential concurrency. In parallel simulation, the research is concentrated on mechanisms that decompose simulations into objects and treat events as atomic actions, which can be executed concurrently. The problem with discrete event simulations is that, to be sure to execute events in increasing simulation time order, logical processes must be prepared to block, before processing an event message, until they can be sure they will never receive another event message with a lower simulation time [Mis86]. In the first section of the paper the basic Time Warp mechanism is described. The next section emphasises our implementation of the algorithm in OCCAM [Occ88]. A number of results, obtained by the introduction of a simple computer network model in the Time Warp frame, are presented in the third section, followed by some optimization proposals. Finally a few conclusions and the future work is outlined.