Process reassignment with reduced migration cost in grid load rebalancing

We study the load rebalancing problem in a heterogeneous grid environment that supports process migration. Given an initial assignment of tasks to machines, the problem consists of finding a process reassignment that achieves a desired better level of load balance with minimum reassignment (process migration) cost. Most previous algorithms for related problems aim mainly at improving the balance level (or makespan) with no explicit concern for the reassignment cost. We propose a heuristic which is based on local search and several optimizing techniques which include the guided local search strategy and the multi-level local search. The searching integrates both the change of workload and the migration cost introduced by a process movement into the movement selection, and enables a good tradeoff between low-cost movements and the improving balance level. Evaluations show that the proposed heuristic can find a solution with much lower migration cost for achieving the same balance level than previous greedy or local search algorithms for a range of problem cases.