Due to the inherent stochastic nature of services execution environment within service oriented systems, a runtime adaptation of the given composition may be required. The authors investigate a runtime service adaptation mechanism based on conditional retries for orchestrated web services. The conditional retry may be issued while a concrete service within composition is executed. The retry could either invoke the same concrete service or a functionally equivalent web service that implements the same task. The authors use dynamic programming to determine the optimal time instances at which the current request should be terminated before request replication. The calculation takes into account different QoS parameters like services' response-time distributions and cost-related parameters, and the solution optimizes the expected revenue of the composite service provider. The authors illustrate the benefits of their approach by numerical calculations, and discuss the impact of considered QoS parameters to the solution at hand.
- Composite service provider (CSP)
- Dynamic programming
- Heavy tailed distributions
- Optimal retry policies
- Service oriented architecture