Performance Comparison Between MPI and AMPI

07/27/2011 - 2:05pm to 2:25pm
Main Seminar Room - ML
Rabin Ranabhat

Rabin Ranabhat, Loras College

Abstract:  While implementing a program in parallel, it is desirable that there is equal load balance between all the processes executing the parallel program. In order to do so, dynamic load balancing is of great help as it avoids significant changes to the design and structure at the application level. However, most MPI implementations do not readily support dynamic load balancing. On the other hand, AMPI efficiently supports dynamic load balancing by using object migration for irregular and dynamic applications and also deal with external factors that cause load imbalance. The goal of this SIParCS project is to consider performance test of AMPI and MPI using a parallel program. While doing so, parallel IO can present some problem to AMPI as object migration for parallel IO is not a possibility. The thing that is considered for a performance test is the computation time taken by each processes. Results and comparisons of performance by MPI and AMPI after load balancing will be discussed in the presentation.

Video replay of the presentation

Please install the Flash Plug-in or click here for non-flash supported devices.

Download the entire presentation (right-click/save link as)

Presented on July 27, 2011 at NCAR