Experiments with the CHIME Parallel Processing System

摘要

This paper presents the results from running five experiments with the Chime Parallel Processing System. The Chime System is an implementation of the CC++ programming language (parallel part) on a network of computers. Chime offers ease of programming, shared memory, fault tolerance, load balancing and the ability to nest parallel computations. The system has performance comparable with most parallel processing environments. The experiments include a performance experiment (to measure Chime overhead), a load balancing experiment (to show even balancing of work between slow and fast machines), a fault tolerance experiment (to show the effects of multiple machine failures), a recursion experiment (to show how programs can use nesting and recursion) and a fine-grain experiment (to show the viability of executions with fine grain computations.