CaDS Seminar 2024 - June 4

Prof. Serge G. Petiton (Univ. Lille & CNRS, France)

Sequences of very sparse and very large matrix computation on Fugaku

Abstract:

Exascale machines are now available, based on several different arithmetic (from 64-bit to 16-32 bit arithmetics, including mixed versions and some that are no longer IEEE compliant) and using different architectures. Recent brain-scale applications, from machine learning and AI for example, manipulate huge graphs or meshes that lead to very sparse nonsymmetrical linear algebra problems. Those applications generate irregularly structured data (graphs, meshes or directly sparse matrices) which allow to distribute along supercomputer nodes a few compressed rows or columns of extreme scale and very sparse matrices but which don’t allow to store any dense vector of the same order on each node, as we usually expected to compute distributed matrix-vector products (even using MapReduce).  In this talk, after a short description of recent evolutions having important impacts on our results, in particular about parallel and distributed iterative methods, I present some results obtained on Fugaku, with Japanese and French colleagues, based on sequences of sparse non-symmetrical matrix products optimized for very irregular sparse and very large matrices. I discuss the performance with respect to the sparsity and the size of the matrices, to some formats to compress the sparse matrices, to the number of process and nodes, and to two different interconnecting network topologies. I also analyze the impact having networks on chip to interconnected some subsets of cores, which don’t share memories, with respect to the sparse irregular patterns of the matrices. I conclude proposing some research perspectives and potential collaborations. 

Short Bio of the speaker: Serge G. Petiton received the B.S. degree in mathematics, the Ph.D. degree in computer science, and the “Habilitation à diriger des recherches”, from the Sorbonne University, Pierre et Marie Curie Campus. He was post-doc student, registered at the graduate school, and junior researcher scientist at Yale University, 1989-1990. He has been researcher at the “Site Experimental en Hyperparallelisme” (supported by CNRS, CEA, and the French DoD) from 1991 to 1994. He also was affiliate research scientist at Yale and visiting research fellow in several US laboratories (NASA/ICASE, AHPCRC,..) during the period 1991-1994. Since 1994, Serge G. Petiton is tenured full Professor at the University of Lille in France and he had a CNRS senior position at the “Maison de la Simulation” in Paris-Saclay, 2013-2021.  Serge G. Petiton was visiting awarded Professor at the Chinese Academy of Science, in 2016. He was P.I. of several international projects with Germany and Japan (ANR, CNRS, SPPEXA,..) and had-have many industrial collaborations (TOTAL, CEA, Airbus, Nvidia, Intel,…).  Serge G. Petiton has been scientific director of more than 30 Ph.D.s and has authored more than 150 articles on international journals, books, and conferences. His main current research interests are in “Parallel and Distributed Computing”, “Sparse Linear Algebra”, “Language and Programming Paradigms”, and “Machine Learning-Transformer methods”.