Error Analysis Techniques Of Performance Measurement Based On Hardware Performance Monitoring

High Performance Computing (HPC) is widely used in kinds of fields of science and engineering to solve large computation problems. With the advance in HPC, many high performance computers are emerged and used. The computers'peak performances increase with a continuous and rapid way. But the sustained performances achieved by real applications do not increase in the same scale as the machine's peak performances do. The gaps between the applications'sustained performances and the machines'peak performances are widening.Program performance optimization is one of the effective ways to solve this problem. Program performance measurement - as the first step of program performance optimization, recognizes the bottleneck of the program performance to find out the utilization ratio of software/hardware resources and the machine's performance by understanding program behaviors. Program performance measurement has taken more important effect throughout the process of performance optimization.Among all sorts of existing program performance measurements nowadays, performance measurement method based on hardware performance monitoring gradually catches the attention of academe and industry with the advantages as follows: the performance data collected at the level of micro-architecture is rich and reliable, interfering is small to the tested system, collecting speed is fast and the overhead of system is small. Analyzing the performance data colleted by this method can not only direct the design of performance measurement software and optimize performance measurement method, but also improve the performances of programs and utilize the resources of the system effectively. However, there are many interference sources which can affect this process and usually lead lots of errors to the performance data. Accordingly, it is necessary to investigate and analyze the data of performance measurement based on hardware performance monitoring.In order to decrease the influence of the interference sources during the test and acquire more reliable and more accurate data, this paper has made a comprehensive investigation on the process of performance measurement based on hardware performance monitoring, including main work as follow:(1) We have investigated the function and bottom implement of the performance monitoring hardware, and deeply studied on the IA-64 processors with more advanced mechanism of hardware performance monitoring than others, including program counter sampling,period statistics,invalidated event address sampling and event restriction.(2) We analyze and validate interference sources or error sources that influence performance measurement based on hardware performance monitoring at all levels, which are...