A model and tool for the analysis of combined data flow and control flow multiprocessors

Improved multiprocessor performance can be attained by architectures that combine data flow and control flow concepts. This type of combined architecture is characterized and several examples of previously proposed architectures are given. To permit the analysis of such systems, COSMIC (Combined Ordering Scheme Model with Isolated Components), was developed. It describes the execution of specific algorithms on multiprocessors and facilitates analysis of their performance. Building upon previous modeling efforts such as Petri nets, COSMIC structures the modeling of a system along several issues including computational and overhead costs due to sequencing of operations, synchronization between operations, and contention for limited resources. This structuring allows us to isolate the performance impact associated with each issue. Finally, studying the performance of a system while executing a specific algorithm gives insight into its performance under realistic operating conditions. The model also allows us to study realistically sized algorithms with ease, especially when they are regularly structured. Results of studies on combined systems indicated that partition size is a major factor in the performance of such systems and that an optimal size may be found for given system parameters