Author

Lang

Date of Award

12-1994

Degree Name

Master of Science in Engineering

Department

Electrical and Computer Engineering

First Advisor

Dr. Harrison W. Greenwood

Second Advisor

Dr. Ajay Gupta

Third Advisor

Dr. Sharon Her

Access Setting

Masters Thesis-Open Access

Abstract

In real-time systems, correctness not only depends on the result of the computation but also on the time at which this result is available. The violation of timing constraints in hard real-time systems can be critical to human life or environment. Therefore, the scheduling algorithm for distributed systems has to allocate tasks to processing nodes so that no timing constraints can be violated. In addition to timing constraints, tasks have precedence and fault-tolerance constraints.

Static scheduling allocates tasks to processing nodes before the tasks are executed. Static scheduling problems are known to be NP-hard [4]. Therefore, heuristic techniques are necessary to find schedules. Evolutionary strategies (ES) have been used to find solutions to NP-hard optimization problems by performing a directed random search in a complex fitness landscape. Recently, ES have been shown to efficiently find low schedule length task allocations in non-real time distributed systems [7].

This thesis shows, ES algorithms can find solutions to the static scheduling problem in real-time distributed systems. The effect of the type of the genetic operators, the populations size, and the fitness function on the efficiency of the ES algorithms were investigated. The ES approach was verified by solving two real-world scheduling problems from [10] and [24]. Solutions were found in less than one hour of CPU time on a Sun SPARC IPC computer.

Share

COinS