Optimizing the Access to Read-Only Data in Grid Computing

Abstract

A fundamental problem of grid computing is the communication overhead. One reason of this overhead is the access to remotely stored data. Caching read-only data is a possible alleviation of the problem. In case of grid computing caching can be optimized by using allocation schemes considering the contents of the caches. Possible ways to achieve such an allocation in a grid are the topic of this paper. The paper proposes to use allocation schemes preferring resources with the required data in their caches. In doing so the hit rate of the caches will be increased and as a consequence the average response time of the jobs and the network load will be reduced. Two new possible allocation approaches are discussed and compared with classical allocation schemes. The performance and the costs of the schemes (when applied to large grids) are evaluated using a simulation environment.