Modeling and Application of Performance Optimization for Parallel Algorithms

  • Jun Feng Wang
  • , Gang Yi Ding
  • , Yu Gang Li
  • , Fu Quan Zhang*
  • *Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

In many large-scale scientific and engineering research fields, simulation computing has played a significant role in its development. This type of computing has the characteristics of high complexity and large computational load, and traditional serial algorithms can no longer meet their computing needs. Parallel processing has become a key technology for solving problems. How to give full play to the performance of highperformance computing systems and design parallel algorithms with optimal performance has become a hot issue in current research. This paper proposes a parallel algorithm performance optimization analysis model to guide the design and optimization of parallel algorithms. This model quantitatively analyzes the design optimization of parallel algorithms from two aspects: parallel systems and parallel algorithms. It provides expressions for the acceleration, efficiency, cost, and other indicators of the algorithm, and points out performance bottlenecks in the algorithm. This provides theoretical guidance for maximizing the performance of parallel systems and optimizing parallel algorithms. Based on this model, this article conducts research and analysis on the parallel optimization design of large matrix multiplication algorithms, and conducts experimental verification, greatly improving the parallel efficiency of matrix multiplication.

Original languageEnglish
Pages (from-to)2123-2136
Number of pages14
JournalJournal of Network Intelligence
Volume10
Issue number4
Publication statusPublished - 2025
Externally publishedYes

Keywords

  • computational intensive applications
  • large matrix multiplication
  • performance optimization analysis models

Fingerprint

Dive into the research topics of 'Modeling and Application of Performance Optimization for Parallel Algorithms'. Together they form a unique fingerprint.

Cite this