This paper presents GRIDHPC, a decentralized environment dedicated to high performance computing. It relies on the reconfigurable multi network protocol RMNP to support data exchange between computing nodes on multi network systems with Ethernet, Infiniband, Myrinet, and on OpenMP for the exploitation of computing resources of multicore CPU.We report on scalability of several parallel iterative schemes of computation combined with GRIDHPC. In particular, the experimental results show that GRIDHPC scales up when combined with asynchronous iterative schemes of computation. KEYWORDSasynchronous iterations, computing environment, grid computing, heterogeneous networks, high performance computing, loosely synchronous applications INTRODUCTIONIn this paper, we present the decentralized environment GRIDHPC dedicated to High Performance Computing (HPC) on grid platforms. The HPC applications that we consider are basically loosely synchronous applications like the solution of numerical simulation problems 1 that present frequent data exchanges between computing nodes. GRIDHPC facilitates the use of large scale distributed systems and the work of programmer.In particular, it uses a limited number of communication operations.The GRIDHPC environment allows data exchange between computing nodes with multi network multi-core configurations. It relies on the Reconfigurable Multi Network Protocol (RMNP) to support data exchange on multi-network systems and on OpenMP 2 for the exploitation of computing resources of multi-core CPU.The protocol of communication RMNP is an extension of the Configurable Transport Protocol (CTP) 3 that makes use of the Cactus framework. 4 RMNP can configure itself automatically and dynamically in function of application requirements like scheme of computation that is implemented, ie, synchronous or asynchronous iterative schemes and elements of context like available network interface cards and network topology by choosing the most appropriate communication network and mode between computing nodes. It can use simultaneously several networks like Ethernet, Infiniband, and Myrinet. These features are particularly important since we consider loosely synchronous applications that present frequent data exchanges between computing nodes. To the best of our knowledge, these features have not been carried out previously on environments or runtime systems in the literature.The remainder of this paper is organized as follows. Related work is presented in Section 2. Section 3 deals with our contribution to support communication in a multi-network context. In particular, the Reconfigurable Multi Network Protocol RMNP is presented. Section 4 presents the architecture and task assignation of the GRIDHPC environment. Parallel programming model is given in Section 5. Computational results with the decentralized environment GRIDHPC for the obstacle problem are displayed and analyzed in Section 6. Section 7 concludes this paper. RELATED WORKThe possibility to consider heterogeneous network resources for HPC application...
HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
This paper aims at presenting Peer-To-Peer HPC a decentralized environment that facilitates the use of heterogeneous multi-cluster platform for loosely synchronous applications. The goal is to exploit all the computing resources (all the available cores of computing nodes) as well as all networks, e.g., Ethernet, Infiniband and Myrinet. Peer-To-Peer HPC functionality relies on a reconfigurable multi network protocol RMNP for controlling multiple network adapters and on OpenMP for the exploitation of all the available cores of computing nodes. We report on efficiency obtained with Grid5000 testbed by combining synchronous and asynchronous iterative schemes of computation with Peer-To-Peer HPC. The experimental results show that our environment scales well.
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.