Abdul Halim
Dept. of CSIS
University of Limerick, Ireland
Using Message Passing Interface(MPI) to Parallelise a Simple Genetic Algorithm(SGA) Running on Beowulf Cluster
The aim of this project was to leverage Message Passing Interface(MPI) API for the parallelisation of a machine learning paradigm running on a distributed memory model infrastructure. This High Performance Computing (HPC) architecture is a Beowulf cluster based on commodity hardware and open source software. MPI defines a set of library routines and environment variables to facilitate execution of tasks in parallel across computing nodes using commu- nication and synchronisation constructs. The Machine Learning paradigm is a Simple Genetic Algorithm (SGA), an adaptive heuristic search and optimisa- tion algorithms based on evolutionary ideas of natural selection and genetics, used for function optimisation in this work. Empirical results are presented on the performance gain using the Beowulf cluster. The findings demonstrate the challenges faced by programmers when trying to balance the communication overhead in MPI and computational complexity when parallelising a program using MPI.