论文部分内容阅读
近几年,性能模型作为一种新的并行系统性能分析方法,得到学术界和工业界的广泛重视.给出了一个开放式性能模型框架结构PMPS(n)并实现了该框架下的一个面向I/O密集型并行应用模型PMPS(3),使用该模型分析了各种NPB程序在PⅣ机群系统上的性能.实验结果表明,对存储密集型应用,PMPS(3)模型与PERC模型预测结果相当;对I/O密集型应用性能的预测,PMPS(3)模型优于PERC模型.进一步分析发现,应用的数据相关、控制相关和操作重叠会影响模型预测结果.实验结果还说明了PMPS(n)性能模型具有很好的扩展性.
In recent years, as a new parallel system performance analysis method, performance model has drawn wide attention from academia and industry. An open performance model framework PMPS (n) is presented and a PMPS (3), an I / O intensive parallel application model, is used to analyze the performance of various NPB programs on the PIV cluster system.The experimental results show that for PMPS (3) and PERC model prediction , And PMPS (3) model is better than PERC model in predicting I / O-intensive application performance.Further analysis found that the data correlation, the control correlation and the operation overlap of the applied data will affect the model prediction results.The experimental results also show that PMPS ( n) The performance model has good scalability.