Memetic algorithms applied to the optimization of workflow compositions

The selection of services of a workflow based on Quality of Service (QoS) attributes is an important issue in service-oriented systems. QoS attributes allow for a better selection process based on non-functional quality criteria such as reliability, availability, and response time. Past research has mostly addressed this problem with optimal methods such as linear programming approaches. Given the nature of service-oriented systems where large numbers of services are available with different QoS values, optimal methods are not suitable and therefore, approximate techniques are necessary. In this paper, we investigate Genetic algorithms and particle swarm optimization for the service selection process. In particular, both methods are combined with an optimal assignment algorithm (Munkres algorithm) in order to achieve higher solution qualities (success ratios) and to form a so called memetic algorithm. Experiments are conducted to investigate the suitability of the approaches and to compare the memetic algorithms with their non-memetic counterparts. The results reveal that the memetic algorithms are very suitable for the application to the workflow selection problem.