Accurate modeling and efficient QoS analysis of scalable adaptive systems under bursty workload

Fulfillment of QoS requirements for systems deployed in the Internet is becoming a must. A widespread characteristic of this kind of systems is that they are usually subject to highly variable and bursty workloads. The allocation of resources to fulfill QoS requirements during the peak workloads could entail a waste of computing resources. A solution is to provide the system with self-adaptive techniques that can allocate resources only when and where they are required. We pursue the QoS evaluation of workload-aware self-adaptive systems based on stochastic models. In particular, this work proposes an accurate modeling of the workload variability and burstiness phenomena based on previous approaches that use Markov Modulated Poisson Processes. We extend these approaches in order to accurately model the variations of the workload strongly influence the QoS of the self-adaptive system. Unfortunately, this stochastic modeling may lead to a non tractable QoS analysis. Consequently, this work also develops an efficient procedure for carrying out the QoS analysis.