MPI Performance


Introduction

We have measured MPI performance on the NOW clusters. This document provides the performance data in both table and graph format.


Performance of MPI on Generic Active Messages - Point to Point


Performance of MPI on Shared Memory SMP - Point to Point


Performance of MPI on Active Messages 2.0 - Point to Point


Performance of MPI on P4 (10 Mbit Ethernet) - Point to Point


System Specification of SUN Ultra Sparc Cluster

System ModelUltra 1
Processor Name167 MHz UltraSPARC
Number of Processor1
L1 Cache Per Processor32 kBytes
L2 Cache Per Processor512 kBytes
Main Memory128 MBytes
OSSUN Solaris 2.5.1
Network InterfaceMyrinet Lanai 4.1 (37.5 MHz) SBus NI card
Number of NI1
Network SwitchMyrinet 8-port Switch

System Specification of SUN Enterprise 5000 SMP

System ModelUltra Enterprise
Processor Name167 MHz UltraSPARC
Number of Processor8
L1 Cache Per ProcessorI-cache: 16 kBytes; D-cache: 16 kBytes
L2 Cache Per ProcessorUnified cache 1 MBytes
Main Memory512 MBytes
OSSUN Solaris 2.5.1
Network InterfaceMyrinet Lanai 4.1 (37.5 MHz) SBus NI card
Number of NI1
Network SwitchMyrinet 8-port Switch
Frederick Wong
Department of Electrical Engineering and Computer Science
University of California, Berkeley
Last Modify: