一种对基于云上的并行应用的性能分析模型
A PERFORMANCE ANALYSIS MODEL FOR CLOUD-BASED PARALLEL APPLICATIONS
陈亮 1陈文智 1张紫徽 1曹学锋 2杨德华2
作者信息
- 1. 浙江大学信息技术中心 浙江杭州 310027
- 2. 杭州数列网络科技有限责任公司 浙江杭州 311100
- 折叠
摘要
由于部署高性能计算相对昂贵,人们利用云计算来作为替代方案.然而,由于资源共享,增加对云访问的处理和网络资源的延迟,导致云环境对长期运行时程序来说变得不可预测.因此,建模和理解性能对于开发这种环境至关重要.该文针对基于消息传递接口的应用程序在云上的执行时间提供了一个预测器.该预测器基于性能分析模型,将云资源视为排队网络,将并行应用程序视为争夺共享资源的任务.实验表明,所提方法的总体准确率为89％.
Abstract
Because deploying high-performance computing is relatively expensive,people use cloud computing as an alternative.However,due to resource sharing,the processing of cloud access and the delay of network resources are increased.As a result,the cloud is an unpredictable environment for long-term runtime programs.Therefore,modeling and understanding performance are critical to developing this environment.This paper provides a predictor for the execution time of an application based on a message passing interface on the cloud.Based on the performance analysis model,the predictor treated cloud resources as a queuing network and parallel applications as tasks competing for shared resources.Experiments show that the overall accuracy of this proposed method is 89％.
关键词
高性能计算/云计算/性能分析/消息传递接口Key words
High-performance computing/Cloud computing/Performance analysis/Message passing interface引用本文复制引用
基金项目
国家自然科学基金项目(61702451)
国家自然科学基金项目(61702452)
出版年
2024
国家科技期刊平台
NSTL
万方数据