Local/Global Control Integration
Abstract
The goal of this contract was to investigate decentralized control with respect to the scheduling and reallocation functions of distributed computer systems. We have made significant progress on developing and analyzing several scheduling and reallocation algorithms. In particular, we have investigated distributed scheduling algorithms where tasks are independent of each other and the subnet imposes a non-negligible delay on task transfers. We have also studied distributed scheduling algorithms that specifically consider collections of related tasks which we classify as distributed scheduling groups and clusters. Since many distributed systems have nodes which are multiprocessors, we have also addressed multiprocessor scheduling. We have studied scheduling in such and environment by analytically analyzing the performance of fork-join jobs and by developing a multi-class, multiprocessor scheduling algorithm. During this contract period, we have also developed a decentralized reallocation algorithm, analyzed it via simulation, and in the analysis, emphasized different forms and costs of cooperation. We show that decentralized reallocation is significantly better than centralized reallocation. Finally, we also developed decentralized estimation techniques to be used in conjunction with distributed scheduling algorithms.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1988
- Accession Number
- ADA201117
Entities
People
- C. G. Rommel
- Don Towsley
- John A. Stankovic
- Ravi Mirchandaney
- Spiridon Pulidas
Organizations
- University of Massachusetts Amherst