Scheduling Support for an Internetwork of Heterogeneous, Autonomous Processors
Steve J. Chapin, Eugene Howard Spafford · Purdue e-Pubs (Purdue University System) · 1991
We are investigating support for distributed, hierarchical scheduling of tasks on autonomous, heterogeneous computing systems. Many researchers have studied the related problem of determining near-optimal task placement in systems possessing some of these attributes. Their algorithms assume the existence of mechanisms to gather information about the system, move tasks, and perform other related operations. We are attempting to define and construct these mechanisms in an abstract fashion that allows them to be generalized to other distributed architectures. Autonomous systems consist of one or more subsystems connected by a message-passing communications mediumj at the lowest level, a processor is an autonomous system with DO subsystems. Processors within the same system may be of different types. All information, behavior, and policy pertaining to an autonomous system is private and local to that system. Any sharing of this information is at the discretion of the local system. The goals of our research are to provide scalable mechanisms for efficient implementations of scheduling policies on systems ranging from a few workstations on a local-area network to thousands of machines spread over a large geographical area and connected by arbitrary interconnection links. We are examining the requirements of these systems for different levels of security, reliability, load sharing, and location transparency. We hope to be able to characterize these and other properties in a scalable mechanism that does not impose undue load demands while supporting a wide range of distributed scheduling policies.