Performance measurement of parallel Ada

Abstract

This paper reports on the development of benchmarks and performance measures for parallel Ada tasking. The focus is on the macroscopic behavior of the benchmarks across a set of load parameters because parallel processing of Ada tasks involves complex run-time behavior and side effects. An Ada program of an application with parallel processes was implemented and its tasks' execution on a multiprocessor system was studied.The chosen application was the NASREM model developed by National Bureau of Standards (NBS). The purpose of the model is to serve as a standard reference control architecture for intelligent, autonomous telerobotic systems. The control architectures of these systems have significant communication requirements as well as computational requirements. A preliminary load model of communication and computation characteristics has been made.Experiments were run on a Sequent Balance 8000 which has a tightly coupled, shared memory multiprocessor architecture and hosts a proprietary version of UNIX. The number of processors varied from 1 to 16. The software environment was a Verdix Ada compiler. A proprietary Ada run-time environment automatically scheduled Ada tasks for parallel execution on available processors.Most results show lowered communication response time as more processors were made available. However, in some cases communication response time increased as more processors were added. This appears because of system overhead.