The Approach to Building the System Software for Parallel Supercomputers

摘要

Presented are the architecture and software for the supercomputers MVS-100, MVS-1000, MVS-1000M. Given is the analysis of causes for another, unlike MVS-1000M, approach to the creation of system software for parallel supercomputers, namely: the implementation of joint use of resources of many, in the general case, various computer systems within the network environment of distributed high-performance computations, e.g. on the basis of the Grid technology.rnPresented is the variant of implementation of system software for parallel supercomputers on the basis of Globus, the original proprietary systems of monitoring, resources and tasks management being built in the Globus.rnGiven are the characteristics and efficiency estimates for the proposed approach to the creation of system software for the next-generation parallel supercomputers.