Job completion prediction in grid using distributed case-based reasoning

Grid allows several entities to share their computational resources. Selecting the best resource to run a job can become a complex and inadequate task for the user since grid is a distributed, dynamic, and heterogeneous network. The current frameworks for this problem still face some challenges. Users never know when the job will finish and what the service provider guarantees. Moreover, job scheduling for a future time is unavailable in most existing framework solutions since they lack performance prediction techniques. This paper presents an approach to job execution time prediction in grid using the case-based reasoning paradigm. The prediction module presented is part of a multi-agent system that selects the best resource to run a job in the grid environment. Case retrieval algorithms involving relevance and geometric matching are presented. We also elaborate adaptation algorithms that use prediction techniques for job workload forecasting.