The effect of strict firing and real characteristics of multiprocessors on performance-a simulation approach

As the complexity of a computing system grows, an analytical model for performance projection becomes more difficult to devise, and inherently contains more simplifying assumptions, which in turn affect the accuracy of the model. The authors' simulation approach is based on a behavioral model for parallel programs. It is behavioral because it assumes the knowledge of the parallel program run-time behavior. A simulator maps (or 'executes') behavioral programs onto a multiprocessor system and produces a set of statistics used in performance evaluation. Preliminary experiments studied the effect of the task firing strategy on program performance. Systems with different characteristics have been investigated.<>