An Evolutionary Approach to Multiprocessor Scheduling of Dependent Tasks

The scheduling of application tasks is a problem that occurs in all multiprocessor systems This problem becomes even more compli cated if the tasks are not independent but are interrelated by mutual exclusion and precedence constraints This paper presents an approach for pre runtime scheduling of periodic tasks on multiple processors for a real time system that must meet hard deadlines The tasks can be related to each other by mutual exclusion and precedence forming an acyclic graph The proposed scheduler is based on Genetic Algorithms which relieves the user from knowing how to construct a solution Consequently the paper focuses on the problem encoding i e the representation of the problem by genes and chromo somes and the derivation of an appropriate tness function The main bene t of the approach is that it is scalable to any number of processors and can easily be extended to incorporate further requirements