Resource allocation on distributed computer systems
Abstract
Distributed computing systems provide cost-effective processing power in response to growing demands. As these systems and the problems attempted become more complex, efficient scheduling techniques are essential to achieve satisfactory performance for users. A goal is to develop a methodology for implementation of a scheduling strategy that uses distributed scheduling intelligence. In this thesis, the results of a theoretical study of scheduling on a bus network are summarized to demonstrate in detail such a scheduling strategy. The strategy uses the existing broadcast capability of multiaccess networks to implement an efficient search technique for finding the extremum of a set of numbers, which reflect the workloads in the computers. For this strategy, the overhead is independent of the number of contending stations which is significant because the primary overhead in scheduling decisions is the collection of status information. Next, the implementation of a global scheduling strategy on a network of Sun workstations is overviewed. This implementation consists of three tasks: developing a workload monitor to dynamically assess the status of participating processors by modifying the Unix operating system, implementing the distributed search to isolate the maximally loaded processor(s), and migrating jobs for remote execution. Finally, areas for future research are identified.
Degree
Ph.D.
Advisors
Wah, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.