Using a classifier system to improve dynamic load balancing
Jan M. Correa, Alba Cristina Magalhães Alves de Melo · 2002
Dynamic load balancing is a very important problem in distributed processing. This problem aims to redistribute running processes to achieve better results according some optimization criterion. Since it is an NP-complete problem in its general formulation, it is worth using heuristics to seek better results in a reasonable time. One of the heuristics that has been successfully applied in various static scheduling problems is genetic algorithms (GAs). We propose to use a classifier system that is an adaptive system that applies a GA over a population of decision rules to achieve better decisions about when to carry out preemptive migrations in a distributed environment. The results have been impressive and the classifier system was able to surpass, without previous knowledge of the workload, the performance of a well designed analytic criterion.