Communications Benchmarks on High-End and Commodity-Class Computers

11/1/2004


Click here to start


Table of Contents

Communications Benchmarks on High-End and Commodity-Class Computers

Outline

DisCo: Technical Progress in 2003-4

High-End Systems Evaluated

Commodity Systems (CSx) Prototype / Evaluation Hardware

Commodity Systems (CSx) II.

Interconnects and Networking

Communications Benchmark PMB and EFF_BW

Communication Benchmarks PMB: Pallas MPI Benchmark Suite (V2.2) and B_EFF

Interconnect Benchmark - EFF_BW

Interconnect Benchmark - Latency

Collective Operations (Time - usec) - as function of no.of CPUs

PingPong Performance

MPI_PingPing Performance

MPI_Sendrcv Performance

MPI_Exchange Performance

MPI_allreduce Performance

MPI_Allgather Performance

MPI_Allgatherv Performance

MPI_Bcast Performance

MPI_All to All Performance

MPI_Reduce Performance

MPI_Reduce_scatter Performance

64-CPU Relative Performance for Allreduce

128-CPU Relative Performance for Allreduce

64-CPU Relative Performance for Allgather

128-CPU Relative Performance for Allgather

64-CPU Performance for all collective operations

Global Arrays

Global Array Benchmark I. GET

Global Array Benchmark II. PUT

Global Array Benchmark III. ACCUMULATE

Author: Martyn F. Guest

Email: m.f.guest@dl.ac.uk