The Research, Design And Implementation On Cluster-based Parallel Computing Environment

A cluster is a group of high-performance computers connected by a high-performance interconnection network or a high-speed LAN to achieve high availability, high performance and high scalability with a single system image. The development of clusters is closely related to their parallel computing environments(PCE). PCE includes resource management module, communication module and collective operation module, etc.This paper is based on the development practices on the Dawning 2000 PVM(DPVM) project.In this paper, we first give an overview of clusters in the following aspects: the development trend, the architecture, and the main features. Then we discuss DPVM's multitudinous advantages provided by Dawning 2000, and explore the details of the design and implementation of DPVM. The performance evaluation is also provided.The emphasis of this paper is the design and the implementation of the PCE in Dawning homogeneous cluster. This PCE is supported by Myrinet highspeed network, BCL(a simplified communication protocol designed by Dawning 2000 project) and Resource Management System(RMS). Myrinet and BCL provide powerful communication capability for DPVM. Thanks to RMS, DPVM is able to be fully compatible with standard PVM in resource management and collective operation. After careful perfomiance evaluation, we can draw the conclusion that DPVM is not only semantically compatible with...