Abstract-Today, clusters are often interconnected by long distance networks to compose grids and to provide users with a huge number of available ressources. To write parallel applications, developers are generally using the standard communication library MPI, which has been optimized for clusters. However, two main features of grids -long distance networks and technological heterogeneity -raise the question of MPI efficiency in grids. This paper presents an evaluation and tuning of four recent MPI implementations (MPICH2, MPICH-Madeleine, OpenMPI and YAMPII) in a research grid: Grid'5000. The comparison is based on the execution of pingpong and NAS Parallel Benchmarks. We show that these implementations present several performance differences. We show that YAMPII performs better results than the others. But we argue that executing MPI applications on a grid can be beneficial if some specific parameters are well tuned. The paper details, for each implementation, the tuning leading the best performances.I. INTRODUCTION Today, clusters are often interconnected by long distance networks within grids to provide users with a huge number of available ressources.Users want to execute their applications, written for clusters, on grids to get better performance by simply using more resources. For example, geophysical applications like ray2mesh [1], a seismic ray tracing in a 3D mesh of the Earth, or medical applications like Simri [2], a 3D Magnetic Resonance Imaging (MRI) simulator, have been executed on grids with quite success and but generally with a sublinear speedup. Indeed, MPI, the standard communication library used to write such parallel applications has been initially implemented and optimized for clusters. Most of implementations used in grids do not consider the specificities of grid interconnections, leading to suboptimal performance.The grid raises mainly three problems to execute MPI applications. First, MPI implementations have to manage efficiently the long-distance between sites. Actually, the high latency between sites is very costly, especially for little messages. Inter-sites communications take more time than intra-sites ones. On an other hand, inter-sites links may offer higher capacities than cluster links.Secondly, MPI implementations have to manage heterogeneity of the different high speed networks composing the grids. Intra-site communications like for example, through a Myrinet interconnect and an Infiniband cluster on the same site and also inter-sites communications through a WAN (Wide Area Network) have to interoperate.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.