Planning (for example choosing most suitable servicesfor self-configuration) is one important task in selfmanagement for pervasive service computing, and can bereduced to the problem of multi-objective services selectionwith constraints. Genetic algorithms (GAs) are effectivein solving such multi-objective optimization problems, andare one of the most successful computational intelligenceapproaches currently available. GAs are beginning to beused in planning for self-management, but there is a lack ofcomprehensive work that evaluates GAs performance andsolution quality, and guides the setting of GAs’ parameters.This situation makes the application of GAs difficultin the pervasive service computing domain in which performance may be critical and the settings of parameters may have big consequences for performance. In this paper, wewill present our evaluations of two GAs, namely NSGA-IIand MOCell, in the GA framework JMetal2.1, for achievingmulti-objective selection of available services. From theseevaluations, suggestions on how and when to use NSGA-IIand MOCell are given in the planning for self-management.Our experiences show that to get a true Pareto front for aproblem, combining solutions set from different GAs is abetter way than using a single GA.
Proceedings of the 14th Ieee International Conference on Enginerring of Complex Computer Systems (iceccs 2009), 2009, p. 192-201
Self-management Planning; Genetic Algorithms; service selection
Main Research Area:
IEEE International Conference on Enginerring of Complex Computer Systems, ICECCS, 2009