Roger W. Hockney

The communication challenge for MPP: Intel Paragon and Meiko CS-2

The communication performance of the Intel iPSC/860, Paragon XP/S and the Meiko CS-2 are compared using the COMMS1 benchmark from the Genesis Parallel Benchmark Suite. The challenge to distributed-memory massively-parallel processors presented by the Cray-C90 shared memory computer is highlighted by re-interpreting vector processing results as though they were measuring communication startup and bandwidth. The results show a wide gap between the two types of computer, in favour of the C-90. These results are for the initial issue of software and hardware of the Paragon and CS-2. Comments from Intel and Meiko are included to show how the manufacturers aim to improve communication performance.

Comparison of communications on the Intel iPSC/860 and Touchstone Delta

Abstract The Touchstone Delta is found to have an asymptotic bandwidth of 6.7 MB/s which is 2.4 times faster than the iPSC/860, but only about a quarter of the advertised rate of 25 MB/s. The Deltas measured startup time of 61 μs is very little less than the iPSC value of 76 μs, however unlike the iPSC, it is independent (within the error of measurement) of the separation between nodes.

Benchmarking for distributed memory parallel systems: gaining insight from numbers

The inherent difficulties of benchmarking parallel computers and the motivation behind the development of the GENESIS benchmark suite introduce the paper. Some of the broader issues that must be taken into account with any rigorous benchmarking effort on a distributed memory system are then discussed together with a relevant example. The initial version of the GENESIS benchmarks was used to evaluate the SUPRENUM computer. The later releases of the suite have been developed to achieve portability of codes across different distributed memory systems by means of the PARMACS macros. Strict portability is however infeasible due to the relative immaturity of current distributed memory computers and the presence of so many different architectures. Therefore some aspects of transportable software are also described, which were felt important when porting the benchmarks to different parallel platforms.

The graphical benchmark information service

Unlike single-processor benchmarks, multiprocessor benchmarks can yield tens of numbers for each benchmark on each computer, as factors such as the number of processors and problem size are varied. A graphical display of performance surfaces therefore provides a satisfactory way of comparing results. The University of Southampton has developed the Graphical Benchmark Information Service (GBIS) on the World Wide Web (WWW) to display interactively graphs of user-selected benchmark results from the GENESIS and PARKBENCH benchmark suites.

Low-level benchmarking: performance profiles

The aim of this paper is to advance the presentation of low-level benchmarking results in the form of performance profiles, which present concisely on one or two pages the single- or multi-processor performance results. A profile consists of measured and derived performance parameters which characterise the performance of parallel computers. In order to achieve a more systematic approach to low-level benchmarking methodology, we also introduce the concept of the benchmarking frame. Several sets of low-level benchmarking results from current RISC platforms including DEC Alpha, SGI Onyx, and SUN Ultra-2 are shown to prove the validity of our approach.

Comparative performance analysis of uniformly distributed applications

A simple programming model of distributed-memory message-passing computer systems is first applied to describe the coupled architecture/application by two sets of parameters. The node timing formula is then derived on the basis of scalar, vector and communication components. A set of suitability functions, extracted from the performance formulae, are defined. These functions are shown to be particularly useful for comparative performance analysis of both existing distributed-memory systems and new architectures under development.<<ETX>>

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

The Genesis benchmark suite has been assembled to evaluate the performance of distributed-memory MIMD systems. The problems selected all have a scientific origin (mostly from physics or theoretical chemistry), and range from synthetic code fragments designed to measure the basic hardware properties of the computer (especially communication and synchronisation overheads), through commonly used library subroutines, to full application codes. This is the second of a series of papers on the Genesis distributed-memory benchmarks, which were developed under the European ESPRIT research program. Results are presented for the SUPRENUM and iPSC/860 computers when running the following benchmarks: COMMS1 (communications), TRANS1 (matrix transpose), FFT1 (fast Fourier transform) and QCD2 (conjugate gradient kernel). 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.