Control-theoretic Adaptation Strategies for Autonomic Reconfigurable Parallel Applications on Cloud Environments

Cloud Computing is a paradigm that enables the access to a set of shared networking and computing resources and high-level platforms and services through the exploitation of virtualization technologies. On Clouds, it is of relevant importance to make applications adaptive and reconfigurable, in the sense that the optimal configuration (satisfying desired QoS levels) should be dynamically changed in response to variations in the workload conditions and in the resource availability. Due to this fact, adaptation strategies have gained much attention over the last years. Properties like control optimality (finding proper trade-offs between contrasting QoS goals), reconfiguration stability (expressed as a function of the average time between consecutive reconfigurations) and reconfiguration amplitude (performing sequences of small modifications of the current configuration) are important aspects to consider. In order to meet these needs, we present a control-theoretic approach and we provide a first validation of our proposals, giving an insight about its applicability to Cloud environments. Keywords—Autonomic Computing, Parallel Computations, Reconfigurations, Model-based Predictive Control, Distributed Cooperative Optimization