Identification and Categorisation of Applications and Initial Benchmarks Suite

This document contains an analysis of a major survey of HPC systems and applications across the PRACE 1 partners (see [1]). The survey data was used to produce an overall utilisation matrix characterised by scientific area and algorithm. This matrix represents one of the best snapshots of HPC utilisation produced, and could be an invaluable basis to predict the likely utilisation of future European Petaflop/s systems. We discuss a methodology for weighting subset lists of applications to maximise their match with the utilisation matrix. This process is used to guide the selection of applications for inclusion in the representative benchmark suite. By selecting highly used applications from a range of scientific areas and algorithms, we were able to produce a list of no more than 16 applications that are generally representative of European HPC usage and fitted the utilisation matrix well. Drawing on the myriad PRACE expertise in existing and emerging applications areas, we produced a recommended list of nine representative applications, based on the best-fitting list, for use in benchmarking future Petaflop/s systems. The future of HPC in Europe is also considered. Programming multi-core architectures is seen as a major challenge alongside parallel I/O and techniques for scaling applications to thousands of cores. By comparing large and small systems in the survey it would appear that Particle Physics may well be the main scientific area in a future Petaflop/s system, alongside Materials and Computational Chemistry. The study detailed in this report is a detailed snapshot of current European HPC, enabling both the choice of a representative benchmark suite and providing insight into what might run on future systems.