Reliable Scientific Service Compositions

Distributed service oriented architectures (SOAs) are increasingly used by users, who are insufficiently skilled in the art of distributed system programming. A good example are computational scientists who build large-scale distributed systems using service-oriented Grid computing infrastructures. Computational scientists use these infrastructure to build scientific applications, which are composed from basic Web services into larger orchestrations using workflow languages, such as the Business Process Execution Language. For these users reliability of the infrastructure is of significant importance and that has to be provided in the presence of hardware or operational failures. The primitives available to achieve such reliability currently leave much to be desired by users who do not necessarily have a strong education in distributed system construction. We characterise scientific service compositions and the environment they operate in by introducing the notion of global scientific BPEL workflows. We outline the threats to the reliability of such workflows and discuss the limited support that available specifications and mechanisms provide to achieve reliability. Furthermore, we propose a line of research to address the identified issues by investigating autonomic mechanisms that assist computational scientists in building, executing and maintaining reliable workflows.