Etude de la Distribution de Calculs Creux sur une Grappe Multi-coeurs

TL;DR

This paper studies the distribution of sparse hollow matrix computations on multi-core clusters, emphasizing parallel data processing for large-scale numerical applications in various scientific fields.

Contribution

It introduces a new analysis of hollow matrix calculations distribution on multi-core clusters, enhancing parallel computation efficiency for large sparse matrices.

Findings

Improved parallel distribution strategies for hollow matrices.

Enhanced computational performance on multi-core architectures.

Better resource utilization in high-performance computing environments.

Abstract

Nowadays, high performance computing is becoming more and more important in different fields research and industry, such as medical imaging and diagnostics, mathematics as well as oil exploration. It refers to intensive computing in some applications where one needs to use a large number of computing resources (computing power, memory rate, storage space, etc.). Thus, it is necessary in this case to run these applications on architectures parallel making multiple computers work together and running over 10 operations at floating point per second (or a petaflops). 15 Parallel computation consists of executing one or more programs, simultaneously, by multiple processors. In general, we have two ways to perform a parallel calculation. The first is to cut the program into several calculation tasks then, run all these parallel spots by different processors. The second requires partitioning the data, so that each part of the data is assigned to a different processor. Then, all the processors run the instructions of the same program in parallel but by operating on different data . This last method, called parallelism of data, is that retained in this memory. Digital analysis is one of the areas where the use of platforms is essential parallel, in particular, in the case where numerical methods treat hollow matrices. This last method, called parallelism of data, is that retained in this memory. Digital analysis is one of the areas where the use of platforms is essential parallel, in particular, in the case where numerical methods treat hollow matrices. A hollow matrix is a very large matrix that contains a small proportion non-zero elements. These matrices can come from different domains such as simulation structural mechanics, image / signal processing, the study of the dynamics of fluids etc.