Flexible Provisioning of Service Workflows

As computer systems are becoming increasingly distributed and complex in nature, services are emerging as a promising paradigm for offering processes and resources on demand to consumers. By relying onmachine-readable service descriptions, software agents in such systems are able to discover and invoke remote services without human intervention at run-time. This allows them to achieve complex goals, often by combining many atomic services into larger workflows. Current work on service-oriented computing has so far concentrated on knowledge representation techniques, reasoning over service descriptions and appropriate communication protocols. Such work typically assumes that service providers are reliable and that they publish accurate and truthful service descriptions. However, this assumption is unrealistic in large distributed systems, such as the Web, large-scale computational Grids or peer-to-peer systems. These systems are inherently dynamic and uncertain — services may fail frequently due to hardware problems, bugs or network failures. Furthermore, the availability of services can change rapidly as providers leave or enter the system, and competition for resources can result in uncertain execution times. Additionally, service providers are increasingly likely to be autonomous agents with their own decision-making mechanisms. This means that they may prioritise service requests or even ignore them if that is in their interest. The resulting uncertainty and unreliability must be addressed when designing software agents that execute complex workflows in service-oriented systems. This issue is particularly critical in scenarios where workflows represent a real value to the consumer, where they must be completed within a limited amount of time, and where service providers demand remuneration. In these cases, any service failures or even delays can jeopardise the overall outcome and result in losses to the consumer. In my thesis, I am addressing these challenges by considering the provisioning of service workflows. During provisioning, the service consumer decides which providers to invoke for the constituent tasks of an abstract workflow. This allows it to choose more reliable providers where necessary