Decentralized Control of Scheduling in Distributed Systems

摘要

A distributed processing system is defined as a collection of processor-memory pairs (hosts) that are physically and logically interconnected, with decentralized system-wide control of all resources, for the cooperative execution of application programs. Such systems may be dedicated to a single application or may implement a general purpose computing facility. By decentralized system-wide control is meant that there exists distributed resources in the system, that there is decentralized control of these resources (i.e., there is no single, central host in charge, nor is there a central state table), that there is system-wide cooperation between independent hosts which results in a single unified system. By system-wide cooperation is meant that the algorithms of the system operate for the good of the whole and not for a particular host. For systems meeting this restrictive definition of distributed processing, it is hypothesized that their reliability, extensibility, and performance will be better than a what is generally available today. In this report the term distributed processing refers to this very specific type of highly integrated distributed system. The major objective of this research project was to develop and compare decentralized scheduling algorithms for distributed processing systems.