Platform Extensions of a Distributed Simulation Kernel

摘要

Since various user supported services are explicitly avoided in the Message Passing Interface (MPI) standard, an implementation of MPI promises to be faster than the Parallel Virtual Machine (PVM), which includes such services by default. However, the MPI implementation LAM/MPI includes such additional services as well, which are comparable with those of PVM. In order to compare the performance of PVM with LAM/MPI in the context of a distributed simulation kernel, we have implemented an MPI version of the Georgia Tech Time Warp (GTW) by replacing its PVM communication with LAM/MPI. In this contribution, we discuss our experiences with the extension of GTW to Linux and LAM/MPI and compare the simulation results.