Process Migration in Grid Computing. Combining Multiple Metrics to Control Process Rescheduling in Response to Resource and Application Dynamics

      Аннотация: This book covers the problem of process migration on grid computing environments. The rescheduling technique is useful for providing dynamic load rebalancing among the available resources. Good migration politics must be used for both improving resource utilization and minimizing the application execution time. Besides discussions about load balancing and scheduling, this book presents a rescheduling model called MigBSP which controls the process migration on Bulk-Synchronous Parallel (BSP) applications. The BSP model was adopted because round-based computations represent a common style for written successfully parallel programs. MigBSP answers the following questions regarding process migration: a) When to launch the rescheduling; b) Which are the candidates for migration; c) Where to put the elected processes for migration. These questions are solved by determining the Potential of Migration (PM) of each process. PM represents the differential idea of MigBSP and combines data from Computation, Communication and Memory metrics. The results show performance improvements in the application time and affirm the migration mechanism as a key technique for platforms like grids.

Ключевые слова: heuristics, High-Performance Computing, Grid Computing, rescheduling, Cluster Computing, load balancing, Process migration