In a distributed system, load balancing or load sharing is a desirable feature that allows a heavily-loaded processor to migrate excess tasks to less busy or idle processors. This paper suggests a technique for locating the appropriate processor for migrated tasks without excessive overhead. In a heterogeneous environment, this location problem is crucial since processors vary and their capability to satisfy a task's requirements may change from time to time. But locating an appropriate processor generally requires considerable communication overhead while state information is gathered about these processors. This paper presents a mechanism called loaded multicasting to locate an appropriate processor to host and execute a migrating task. Multicasting is the key since it provides selectivity while minimizing communication costs. Modeling and simulation experiments are being attempted and prototyping should subsequently follow, but results are still not available. Despite the absence of such results, this paper discusses the approach through examples of load sharing and load balancing algorithms augmented with loaded multicasting.
Keywords: distributed scheduling, distributed systems, load balancing, load distribution, load sharing, multicasting, resource management.