A Fault-Tolerance Pattern Recommender for Transactional Service Compositions

Combining services to build compositions is a key concept in Service Oriented Architecture. In a service composition with transactional property, each service failure may impose a rollback cost on the composition due to the rollback of its dependent successfully ended services. Reducing the mean roll-back cost of a composition is possible by applying Fault Tolerance (FT) pat-terns to each task in order to decrease the failure probability of the composition. Different FT patterns with different imposed cost and execution time overheads can be applied to each task within the composition workflow. In this paper we have formulated the problem of choosing FT patterns for a given transactional composition as an optimization problem and a NSGA-II genetic algorithm has been proposed to find a mapping of patterns to tasks which satisfies conflicting objectives. The effectiveness of the proposed approach has been evaluated by applying the optimization process to different compositions.