A Comparison of MPI and Process Networks

The Message Passing Interface (MPI) is a popular communication library that supports the SIMD model of parallel computing. Process networks (PN), where processes communicate through first-in-first-out channels, support the MIMD model of parallel computing. Java implementations of both MPI and PN are available. We show that the communication primitives available in MPI, such as send/receive, scatter/gather, and broadcast can be emulated in PN. We present benchmark results comparing the performance of mpiJava and to own pn library.

• Type: Conference
• Year: 2005
• Where: IPPS
• Authors: Thomas M. Parks
• Keywords: Distributed And Parallel Computing, IPPS 2005, Message Passing Interface, Parallel Computing, Popular Communication Library