Dynamic Load Balancing Mechanism for Iterative MPI Solvers

Abstract

Message Passing Interface (MPI) remains a dominant model in high per- formance computing today. Load balancing in MPI programs is a key factor that should be considered in order to obtain high performance for parallel computing. The research, Dynamic Load Balancing Mechanism for Iterative MPI Solvers, is conducted in order to gain performance improvement for iterative MPI solvers that is executed on a MPI cluster of heterogeneous multi-core machines by using a dynamic load balancing mechanism. The proposed load balancing mechanism has the capability of executing the given parallel computation in less execution time than the typical execution scenario. The major factors that are considered for the performance improvement are determining the optimum number of processes for the given iterative MPI solver by considering the available resources, work shar- ing, work stealing, process binding to processor cores and execution ow control of MPI processes at the beginning and while the program execution is in progress.