A service composition environment is one where a service interacts with other services to form a composite application. In such an environment, the reliability of an individual service is influenced by the reliability figures of the services it interacts with. Reliability measuring procedures in the past tend to ignore this influence and calculate the individual reliabilities of services in isolation. In this work, we present a technique which allows us to calculate the reliability of a service incorporating the influence of the interacting services. This is done by representing the entire service domain as a continuous time Markov chain and finding the 'failure distance' of each service. The failure distance of a service is an expression of its reliability. The proposed technique is validated by running experiments in a simulated environment wherein services are randomly made to fail and service compositions are formed based on reliability figures of individual services. It is shown that the compositions formed utilizing the proposed technique are much more reliable than those formed with service reliability being calculated in isolation.
展开▼
