Genetic allgorithm for dynamic task scheduling

Abstract

A genetic dynamic scheduling algorithm (GDSA) is applied to the problem of dynamically scheduling multiple tasks. These independent, non- identical tasks must be distributed in a multiprocessor, shared memory system. This problem is known to be NP-complete. The use of a genetic algorithm for dynamic task scheduling offers several advantages to alternative algorithms. A genetic algorithm fids a solution immediately, at initialization. All processing time is devoted to improvement of the solution. This makes them well suited for dynamic applications, where the amount of processing time is likely to vary. Genetic algorithms tend to make the greatest improvements during initial generations. They move quickly to a good solution, although later improvements will become increasingly less fresuent and less dramatic. Schedule makespans using the GDSA are compared to makespans when a first-come, fxst-serve dynamic scheduling approach is used. This opportunistic scheduling algorithm places arriving tasks in a FIFO queue, and pmessors remove tasks from the queue as they become available. Tests were run to schedule 30 and 200 tasks. When 30 tasks were scheduled, the GDSA caused small, but consistent schedule improvements. The GDSA was most effective if the execution times of the tasks were uniformly distributed. However, schedule improvements were also noted with normally and exponentially distributed task execution times. When 200 tasks were scheduled, the GDSA was most effective if the execution times of the tasks were exponentially distributed.