Research On Performance Parameters And Speed Of The Computer Systems

Speed is the most basic parameter of computer performance, for which some methods werepresented. A Multi-core system provides a convenience for the research and experiment of parallelcomputing, and has become the mainstream in the design of system architecture. The integration ofdual-core quad-core and even many-core is contribute to promote the computer system performanceimprovement. However, the increasing number of core brings new problems: if the number ofnuclear increases linearly could lead to a performance increase. In multi-core environment, theutilization of core is one of the factors that promotes computer performance. Therefore, to developthe potential of each nuclear has practical significance to the development of parallel computing.For that reason, this paper analyzes the years of data of TOP500(High performance computerrankings) to research that how to make full use of each core and improve the parallelism bycomputer parameter. The main part works as follows:(1)This paper summarizes the selection of existing computer system performance parametersand test methods, then it analyzes one-sidedness of existing parameter in multi-core systemenvironment and defines a new parameter named core-contribution.(2)This paper expounds the significance of multi-core parallel programming to improvecore-contribution rate and increase speed, besides it analyzes all kinds of parallel algorithm ofparallel program design patterns, on the basis of that it proposes a multi-core parallel programdevelopment method based on the model and does multi-core programming with the thinking ofmodel.(3)It designs a multi-core parallel algorithm solving the problem of Monte Carlo probability，and the task is decomposed into smaller subtasks which enables the parallel computing isadequately, what’s more, it records and analyzes the core-contribution rate the effect of performanceimprovement.(4)By the research of topic crawler it designs a new multi-core parallel algorithm. The problemis converted into a sequential process to analyze the part which can be parallel by the theory ofpipeline, and each task is executed by a special work thread to improve computer performance.