A Performance Evaluation for Fine- . . .

Thread migration is established as a mechanism for achieving dynamic load sharing and data locality. However, migration has not been used with fine-grained parallelism due to the relatively high overheads associated with thread and messaging packages. This paper describes a high performance thread migration system for fine-grained parallelism, implemented with user level threads and user level messages. The thread system supports an extensible event mechanism which permits an efficient interface between the thread and messaging systems without compromising the modularity of either. Migration is supported by user level primitives; applications may implement different migration policies on top of the migration interface provided. The system is portable and can be used directly by application and library writers or serve as a compilation target for parallel programming languages. Detailed performance metrics are presented to evaluate the system. The system runs on a cluster of S..