Based On The Cmp Clusters Research And Application Of Parallel Programming Model

With the further improvement of basic frequency and chip design complity, the single processor system can not meet the need of processor industry's quick development. In order to meet the more big and complex needs of application, it is the time to introduce an innovative way to design processor. The CMP (chip-on-multiprocessor) multi-core technologies become now processor's mainstream application. To clustering these CMP systems together is a new parallel system better than the limits of a single system. The cluster system's performance improved, however the CMP cluster's parallel programming model research is relative lag. In order to find a good model that is suitable to CMP, massive experiments and the analysis are needed. In view of this question, this article has done the following research work:This article briefly introduce the multi-core technology development, parallel programming model research and the multi-core-processor hardware-related technology, including Simultaneous Multithreading(SMT), Chip-on-MultiProcessor(CMP).This article also introduce in detailed the CMP multi-core technology about the question which between the architecture, the sharing and synchronized technology, operating system design, software design, security and so on. Through experimental verification single PC under multi-core performance lifting efficiency, and we has made the comparative analysis with the traditional SMP cluster. Several common parallel programming models and the latest developments are introduced too.And then we described the hybrid TBBMPI programming model suited for CMP cluster in detail. It is closer to the architecture of SMP cluster, could obtain the better performance. Based on this, we study this model to realize the mechanism with emphasis, the parallelization granularity, the thread number determination, superior measure and so on. Through the performance analysis comparison, we obtain that:the TBBMPI programming model is most suitable for the CMP cluster under the controlled condition.Finally we describe the parallel program design process and experiment's analysis method in detail through a concrete application case-EZW, as well as experiment environment peaceful assembly and so on. The result analysis has demonstrated the work's validity.The appearance of multi-core architecture greatly speeds up the development of CMP Cluster. The rising complex architecture has already turned into the most cost-effective solution for the large scale scientific research applications and the server of for enterprise. This paper views the program model and optimization of the CMP Cluster.