Mapping Algorithms Optimizing the Overall Manhattan Distance for Pre-Occupied Cluster Computers in SLA-Based Grid Environments

摘要

Grid applications are more and more widely used nowadays. One of the major challenges is to provide a reliable and predictable platform for computations of various kinds. In order to overcome this challenge, Grid management systems such as the virtual resource manager (VRM) implement scheduling and mapping algorithms at level of the local management systems with support for resource reservation in advance. In this paper, we examine three different mapping algorithms for supercomputers and cluster systems with respect to execution time and the achieve able performance regarding important metrics such as overall Manhattan distance and achievable utilization. The results show the importance of carefully implementing scheduling and mapping algorithms in Grid environments.