Single-objective versus multi-objective Genetic algorithms for workflow composition based on service level agreements

Workflow composition is a very important issue for service-oriented environments. In particular, the composition of services based on quality of service (QoS) attributes has been gaining attention. The user can describe the request of a workflow in terms of QoS attributes, i.e., the user aims for specific service performance, e.g. low waiting time, high reliability and availability, which are based on different service level plans provided by the providers. Past research has addressed this workflow composition problem classifying them into single-objective and multi-objective optimization solutions. Most of the research has employed the single-objective approach whereby the different objectives, i.e., QoS attributes, are aggregated by a weighted approach. Fewer research approaches have used the multi-objective approach, whereby several solutions are produced by a set of Pareto solutions that have equivalent quality to satisfy specific service level agreements. However, no comparison has been done investigating both approaches. Therefore, this paper addresses this shortcoming by an analysis of both, measuring the success ratio as well as the execution time of two Genetic algorithm implementations (single-objective and multi-objective) applied to the workflow composition problem.