Parallel logic simulation on distributed memory multiprocessors classification and evaluation of different approaches
Peter Luksch · 1995
A test environment is presented that allows for different methods of executing discrete event simulations in parallel to be evaluated in a uniform environment. A great variety of parallelizations have been proposed in the past. Up to now, however, an unbiased comparative evaluation of different approaches has been impossible because run-time measurements published in the literature have been obtained with different simulators on different multiprocessor systems and therefore cannot be compared. Our approach to an unbiased comparison of different parallelization methods is as follows: The variety of existing methods is structured by classifying them according to the way how the simulation task is subdivided into processes and how these processes synchronize. Thus a small number of fundamentally different approaches to distributed simulation can be identified each of which comprises a whole class of parallel algorithms. As a basis for the testbed, one representative from each approach has been implemented. Thus a comparison of different approaches is possible while at the same time a library of functions is provided that allows further parallelizations to be implemented easily.