The Genesis distributed‐memory benchmarks. Part 1: Methodology and general relativity benchmark with results for the SUPRENUM computer

Abstract

SUMMARYThis is the first of a series of papers on the Genesis distributed-memory benchmarks, which were developed under the European ESPRIT research program. The benchmarks provide a standard reference Fortran77 uniprocessor version, a distributed memory MIMD version, and in some cascs a Fortran90 vcrsion suitable for SIMD computers. The problems selected all have a scientific origin (mostly from physics or theoretical chemistry), and range from synthetic code fragments dcsigncd to mcasurc the basic hardware properties of the computer (especially communication and synchronisation overheads), through commonly used library submutines, to full application codes. This first paper defines the methodology to be used to analyse the benchmark results, and gives an example of a fully analysed application benchmark from General Relativity (CR1). First, suitable absolute performance metrics are carefully defined, then thc pcrformance analysis treats the execution time and absolute pcrformance as functions of at least two variables, namely the problem size and the number of proecssors. The theoretical predictions are compared with, or fitted to, the measured results, and then used to predict (with due caution) how the performance might scale for larger problems and more processors than were actually available during the benchmarking. Benchmark measurements are given primarily for the Geman SUPRENUM computer, but also for the IBM 30835, Convcx C210 and a Parsys Supernode with 32 T800-20 transputers.

Efficiency measurements in heterogeneous distributed computing systems: from theory to practice

“…This is the reason why a benchmark rarely consists of a single program, but rather is a suite of codes able to cover a range of different computations [14,15]. On the other hand, efficiency is to be evaluated for the given problem, not for a set of typical problems.…”

“…Elapsed time is defined as the (physical) time spent while the program is running, and is to be measured as if by a clock on the wall. It is customary to measure elapsed time while the program is the sole user of the target computer system, in order to get rid of the CPU and network workload belonging to other users [14].…”

A unified symbolic evaluation framework for parallelizing compilers

“…If there are equalities in a given set of constraints, then they are solved for a speci c variable included in the equality and substituted in the set of constraints. All techniques target wide classes of linear and non-linear symbolic expressions 1 Mathematica and MathLink are registered trademarks of Wolfram Research, Inc. and constraints. For detailed information about our manipulation and simpli cation techniques, which goes beyond the scope of this paper, the reader may refer to 23,24,25,45].…”

The genesis distributed memory benchmarks. Part 2: COMMS1, TRANS1, FFT1 and QCD2 benchmarks on the suprenum and IPSC/860 computers

“…This is the second of a series of papers on the Genesis distributed-memory benchmarks, which have been used for the evaluation of message-passing MIMD computers under the European ESPRIT research program [2,3]. The first paper [4] described the methodology to be used in the analysis of benchmark measurements, and gave results for the general relativity (GR1) benchmark. In this paper we include further results for the SUPRENUM[S] computer and Intel iPSC/860 on the communications (COMMSl), transpose (TRANSl), fast Fourier transform (FFTl) and conjugate gradient (QCD2) benchmarks.…”

“…COMMS 1, see below, for communication), through typical scientific library subroutines (such as the fast Fourier transform FFTl, see below), to stripped down real application codes (e.g. general relativity GRl [4] and a 2-dimensional particle-in-cell code LPMl [6]). This hierarchical approach is common also to the Euroben initiative [7,8].…”