An Experimental Comparison of Initial Placement vs. Process Migration for Load Balancing Strategies

Abstract:

In this paper we describe an experimental comparison between initial placement and process migration to determine which is the better method to use to implement a load balancing system. A trace-based synthetic workload is described, which allows us to control the role that implementation overhead plays in such a system, and the two methods are compared for varying values of this overhead, from significant to negligible.