A Reverse Order-Based QoS Constraint Correction Approach for Optimizing Execution Path for Service Composition

Abstract

In service oriented computing systems, a business process can be exposed as a composite service which consists of a set of logically connected sub-services. For each service in the composition, many service providers can offer the same function but may different QoS. In the general service composition, when a user submits a request, overall QoS constraints called end-to-end QoS composition’s requirements, for example, time should be less than one hour, and cost should be less than 60$, can be transmitted at the same time. As such, how to effectively coordinate individual QoS constraints for single service to achieve the best overall QoS benefits without violating such end-to-end QoS constraint requirements has been a critical issue. With an increasing number of abstract services in a service composition, the possibility of execution path by selecting different service providers for each abstract service blows up exponentially. Therefore, service selection problem for service composition is a computational-hard problem, which can be regarded as a Multiple choice Multiple dimension Knapsack Problem (MMKP) that has been proved np-hard [1, 2, 3]. Recently, a lot of approaches such as graph-based techniques[4], runtime adaptation-based techniques[5], Service Level Agreement(SLA), negotiation and auction based techniques[6], Integer Linear Programming (ILP) based techniques[7] have been proposed to resolve overall QoS constraints for optimizing execution path in a service composition. No matter what the merits and the importance current existing methods have, they rely on directly judging constraint conditions to detect multiple paths for picking out a critical execution path, which easily produces a high-time complexity and even an unsatisfactory result in comparison to the best path. As such, the issue on resolving overall QoS constraints to achieve an optimal execution path has not yet been well addressed.

This paper is supported by Swinburne Dean’s Collaborative Grants Scheme 2007-2008, and by Swinburne Research Development Scheme 2008, and by the National “973” Research Plan Foundation of China under Grant No. 2003CB317008.