Predicting the limits of multiple processor performance using job profiles

Abstract

The authors study the limitations placed on performance by using a model known as the job profile model. They extend and generalize the job profile model to develop a performance model that can be used to determine a lower bound on performance for a class of scheduling policies for a specific architectural configuration. The extended model incorporates the effects of communication between subtasks of a job, and a contention for common resources into the job profile model. This new model provides performance equations that depict the performance impact of the synchronization demands and the algorithmic decomposition of a workload. The extended model is then used to study the effect of the communication to computation ratio on performance. The validity of the model was verified by comparing the theoretical results obtained from the model to experimental results obtained from simulation.<>