Evaluating Budget of Overhead and Scalability on High-Performance Computing Systems

High-Performance Computing Systems (HPCS) based onparallel processing have the potential for satisfying the rapid growthin computational requirements at economical costs. As such machinesgrow in size, however, parallelization overhead grows in a mannerthat can limit scalability. This work sheds some light on the sources,dynamics, and magnitudes of the different types of overhead and theirimpact on performance. Results are obtained through experimentalmeasurements of one of the NASA Earth and Space Sciences (ESS)applications running on the Jet Propulsion Laboratory/Earth andSpace Sciences High-Performance Computing (HPC) systems.