Arrival and delay curve estimation for SLA Calculus

An algorithm and selection method to estimate Network Calculus arrival bounds for systems with concurrent arrivals is presented. Concurrent job arrivals are common for Service-Oriented Architectures. Their performance is described in Service Level Agreements including quantitative requirements for load and response times. SLA Calculus, a variant of Network Calculus, can be used for service performance modeling and validation with SLAs. Functions called curves are used to bound job arrivals as well as their delay. Due to the concurrent nature of job arrivals curve estimation methods used for successive packet arrivals in Network Calculus cannot be applied in SLA Calculus. We present a method to estimate unknown SLA Calculus arrival and delay bounds from input and output traces. This paper introduces an algorithm for the estimation of the curves. Optimal selection of a curve model based on several fitting criteria is performed using candidates from trace sets.