OBSERVATIONS ON USING PROBLEM-SPECIFIC GENETIC ALGORITHM FOR MULTIPROCESSOR REAL-TIME TASK SCHEDULING

Task scheduling is crucial to the performance improvement of multiprocessor systems. Genetic algorithms are extensively used to deal with task scheduling due to the computational intractability of such issues. However, since the genetic algorithm aims to be a generic solution to a variety of problem types, it hardly exploits problem-specific search techniques which might help speed up the search or lead to a better solution. That may compromise the potential power of the genetic algorithms a lot. To overcome this, a problem-specific genetic algorithm is proposed to handle multiprocessor real-time task scheduling in this paper. Rather than only employing limited problem-specific information, our proposal makes the most of such information throughout the evolution of the genetic algorithm. The simulation results show that the performance of the genetic algorithm is greatly improved with the help of certain problem-specific knowledge.