Load Sharing in Distributed Systems

A distributed system consists of, possibly heterogeneous, computing nodes connected by a communication network. Such a system can be used eeectively only if the software presents a single system image of this physically distributed system to its users. Thus all resources of any node should be easily and transparently accessible from any other node. One of the most important of such resources is the CPU. The CPUs of all the nodes in the system can be made transparently available to all nodes if the nodes share their computing load. Thus the system should decide the best node to execute any job regardless of where the job originated and may even migrate some jobs during their execution. This calls for transparent process migration among the nodes. In this paper, we brieey discuss the issues in such load sharing. There are two orthogonal kinds of issues to be addressed here. The rst kind relate to the policies for migration. For example, where should a new job be executed, when and where should an executing process be migrated etc. The second kind of issues relate to the mechanisms for migration. For example, how to checkpoint and transfer the state of a running process, can a process be migrated to a machine with a diierent architecture, can old programs be migrated without modiications etc. We rst brieey survey the current state of the art in this area. Then, we talk about the work being done at IIT Kanpur in this area.