Cost-Sensitive Concurrent Planning Under Duration Uncertainty for Service-Level Agreements

This paper brings together work in stochastic modelling, using the process algebra PEPA, and work in automated planning. Stochastic modelling has been concerned with verification of system performance metrics for some time: given a model of a system, determining whether it will meet a service-level agreement (SLA). For example, whether a given sequence of transitions on a network will complete within 5 seconds 80% of the time. The problem of deciding how to reconfigure the system most cost-effectively when the SLA cannot be met has not been widely explored: it is currently solved manually. Inspired by this, we consider how planning can be used to automate the configuration of service-oriented systems. Configuring these stochastic systems presents new challenges to planning: building plans that meet SLAs, but also have low cost. To this end, we present a domainindependent planner for planning problems with action costs and stochastic durations, and show how this can be used to solve both traditional planning domains, and within the framework of configuring a larger process algebra model.