Dependable composition of transactional web services using fault-tolerance patterns and service scheduling

Building composite value-added services by combining basic services has become a prevalent way of software development. To provide the dependability in a composite service it is essential to combine the backward and forward recovery strategies properly based on composition level quality of service (QoS) factors or user preferences. The backward recovery, known as a fault-tolerance (FT) technique, is performed by compensating the successfully terminated services dependent on the failed service which may entail rollback costs. The forward recovery is achieved by applying FT patterns to each workflow task to decrease its failure rate. It has the disadvantages of longer invocation time and cost. Obviously the success of these methods in finding transactional compositions is highly dependent on the compensateability and retriability of services. In this study, four composition-level QoS factors have been defined to select between the forward or backward recovery strategies in the workflow. The problem of creating a dependable composite service is formulated as a multiobjective optimisation algorithm which explores not only the huge search space of different recovery strategies but also the cross-cutting search space of task scheduling. The effect of task scheduling on finding solutions with lower costs has been verified using case studies.