Sensitive GRASP: combinatorial optimisation of composite web services guided by sensitivity analysis

Abstract

The task of designing and tuning computer systems for optimal performance and dependability is complex. This usually involves the choice of manufacturers, service providers, types of components, and possible architectures. Analytical modelling is helpful for planning and predicting the quality-of-service QoS measures of a system, and can be especially useful when integrated with optimisation algorithms. This study proposes a sensitive greedy randomised adaptive search procedure Sensitive GRASP, which uses sensitivity analysis of an analytical model to improve the solutions of a GRASP-based optimisation. Our approach is applied for optimising the reliability and performance of a composite web service. For scenarios with thousands of possible combinations, the proposed approach finds configurations that are closer to the optimal solution than those found by the algorithm without sensitivity analysis with equivalent computational effort.