Home         Authors   Papers   Year of conference   Themes   Organizations        To MES conference

MPI-based Software Model for Parallel Computing in Heterogeneous Clusters
Authors
	Macharadze G.T.
Date of publication
	2021
DOI
	10.31114/2078-7707-2021-3-106-111
Abstract
	The construction of clusters based on heterogeneous computing systems requires the use of special software tools that ensure the effective execution of application software. The paper considers a software model based on MPI, OpenCL, the Linux OS driver, as well as mpi-run service processes. The service process within the computing node generates MPI computing processes using OpenCL. Interaction between processes is carried out by sending messages: processes within one computing node – through shared memory, processes of different computing nodes – through the network socket mechanism. The experimental study of the model was carried out on the Russian heterogeneous SoC RoboDeus using sixteen high-performance ELcore-50 cores developed by JSC ELVEES research center The obtained results of the BLAS functions, as well as the HPL test for solving the problem with dense matrices in the double-precision floating-point format, confirmed the effectiveness of the proposed MPI model.
Keywords
	cluster computing systems, parallel computing, heterogeneous computing systems, problem-oriented cores, processor 1892VM248, RoboDeus, MPI, OpenCL, BLAS, HPL.
Library reference
	Macharadze G.T. MPI-based Software Model for Parallel Computing in Heterogeneous Clusters // Problems of Perspective Micro- and Nanoelectronic Systems Development - 2021. Issue 3. P. 106-111. doi:10.31114/2078-7707-2021-3-106-111
URL of paper
	http://www.mes-conference.ru/data/year2021/pdf/D060.pdf