Design and Implementation of Prophesy Automatic Instrumentation and Data Entry System

In this paper, we present the Prophesy Automatic Instrumentation and Data Entry (PAIDE) system and describe its design framework and implementation supporting C, Fortran77 and Fortran90 programs on diverse systems. We use NAS Conjugate Gradient (CG) and Integer Sort (IS) benchmarks to analyze the PAIDE’s instrumentation overheads for two granularities of instrumentation with increasing number of processors and problem size. The experimental results on the SGI Origin2000 show that the instrumentation overhead for CG benchmarks is less than 3.4%, while that for parallel IS benchmarks is less than 1%. This is significant and it indicates that the PAIDE system does not perturb the performance data. The PAIDE system is able to capture performance information on repeated events as sequential events and is appropriate for parallel and distributed applications that take a long time to execute.