Efficient Parallel Matrix Multiplication on Distributed Memory Architectures Using Symmetric Block Cyclic Allocation

International Journal of Computer Science and Technology | Vol. 5, No. 7, July 2003 | pp. 49–56

DOI: 10.46882/2003/IJCST/000047

Article Type: Research Article

Title: Efficient Parallel Matrix Multiplication on Distributed Memory Architectures Using Symmetric Block Cyclic Allocation

Names of Authors: Ming-Wei Chen¹, Sarah Jenkins²

Authors’ Affiliations: ¹Department of Software Engineering, Tsinghua University, Beijing, China; ²Department of Cybersecurity, MIT, Massachusetts, USA

Abstract: Large-scale scientific computing and engineering simulations depend heavily on the performance of fundamental linear algebra operations executed over distributed clusters. This research presents an optimized parallel matrix multiplication algorithm designed for high-performance distributed memory architectures. We present a Symmetric Block Cyclic Data Allocation (SBCDA) framework that optimizes communication topologies by matching sub-matrix data dimensions with physical processor grid layouts. Message Passing Interface (MPI) primitives are used to coordinate non-blocking communications during parallel execution phases. Performance benchmarks conducted on a cluster of 64 Intel Xeon processors demonstrated a sustained computation speed of 245 GFLOPS for dense 10,000 × 10,000 matrices. This matches an aggregate parallel efficiency of 88.5% relative to theoretical peak processing ceilings. Our balancing technique minimizes communication overhead by 23.4%, offering scalable performance profiles for complex computational fluid dynamics solvers.

Keywords: Parallel Computing, Distributed Memory, Matrix Multiplication, MPI, Load Balancing, Cluster Architecture

Manuscript Timeline: Received: March 10, 2003; Revised: May 02, 2003; Accepted: June 04, 2003; Published: July 11, 2003

Citation: Chen, M.-W., & Jenkins, S. (2003). Efficient Parallel Matrix Multiplication on Distributed Memory Architectures Using Symmetric Block Cyclic Allocation. International Journal of Computer Science and Technology, 5(7), 49–56.

Subscribe to read the full article: https://www.internationalscholarsjournals.org/subscribe-to-read