基于马尔可夫过程的超级计算系统弹性模型  

Resilience Modeling of Supercomputing Systems Based on Markov Processes

作　　者：杨晓东 杨剑锋 邓周灰 何清龙 YANG Xiaodong;YANG Jianfeng;DENG Zhouhui;HE Qinglong(School of Mathematics and Statistics,Guizhou University,Guiyang 550025,China;不详)

机构地区：[1]贵州大学数学与统计学院,贵州贵阳550025 [2]贵大·贵安科创超级计算算力算法应用实验室,贵州贵阳550025 [3]南宁师范大学广西应用数学中心,广西南宁530100 [4]贵安新区科创产业发展有限公司,贵州贵阳550000

出　　处：《武汉理工大学学报(信息与管理工程版)》2024年第6期951-957,共7页Journal of Wuhan University of Technology:Information & Management Engineering

基　　金：国家自然科学基金项目(72361008);贵州省科技计划项目(黔科合支撑[2023]一般268)。

摘　　要：超级计算系统的弹性建模旨在评估和增强系统在面对硬件故障、软件错误、网络中断等多种破坏性事件时的抵抗、吸收和恢复能力。针对主流的分布式并行超级计算系统,建立了基于马尔可夫过程的多状态多节点超级计算系统弹性模型。该模型利用马尔可夫过程描述系统和节点的性能状态转移,针对串并联混合结构,分析系统和节点在不同性能状态随时间变化的弹性情况。通过一组模拟仿真数据,说明了该弹性模型具有较强的有效性和实用性,为超级计算系统的设计、优化和维护提供理论基础。The resilience modeling of supercomputing systems aims to assess and enhance the system′s ability to resist,absorb,and recover from various disruptive events such as hardware failures,software errors,and network interruptions.This paper established a multi-state,multi-node resilience model for mainstream distributed parallel supercomputing systems based on Markov processes.The model used Markov processes to describe the transitions of system and nodes performance states and analyzed the resilience of the system and nodes over time in a series-parallel hybrid structure across different performance states.A set of simulation data demonstrated the effectiveness and practicality of the proposed resilience model,providing a theoretical foundation for the design,optimization,and maintenance of supercomputing systems.

关 键 词：超级计算系统 分布式并行结构 弹性模型 马尔可夫过程 数值仿真 

分 类 号：TB1[一般工业技术]