An Approach for Efficient Execution of SPMD Applications on Multicore Clusters

© 2017 by John Wiley & Sons, Inc. All rights reserved. This chapter describes an efficient execution methodology for multicore clusters, which is based on achieving a suitable application execution with a maximum speedup achievable while the efficiency is maintained over a defined threshold. This methodology allows people to calculate the maximum number of cores that maintain strong application scalability while sustaining a desired efficiency for single program, multiple data (SPMD) applications. The chapter presents a methodology for efficient execution of SPMD applications on multicore clusters. This methodology is focused on managing the communications heterogeneities present on multicore clusters with the objective of improving both efficiency and speedup. This improvement process is realized through four phases: characterization, tile distribution model, mapping strategy and scheduling policy. The chapter explains some practical examples with the aim of validating this methodology. It gives the following applications: heat transfer app, Laplace application and application integrated in the MP-Labs suite (LL-2D-STD-MPI).