Towards Autonomic Cost-Aware Allocation of Cloud Resources

While clouds conceptually facilitate very fine-grai ned resource provisioning, information systems that are able to fully leverage this potential rema in an open research problem. This is due to factors such as significant reconfiguration lead-ti mes and non-trivial dependencies between software and hardware resources. In this work we ad dress these factors explicitly and introduce an accurate workload forecasting model, based on Fo urier Transformation and stochastic processes, paired with an adaptive provisioning fra mework. By automatically identifying the key characteristics in the workload process and estimat ing the residual variation, our model forecasts the workload process in the near future with very h igh accuracy. Our preliminary experimental evaluation results show great promise. When evaluat ed empirically on a real Wikipedia trace our resource provisioning framework successfully utiliz es the workload forecast module to achieve superior resource utilization efficiency under cons tant service level objective satisfaction. More generally, this work corroborates the potential of holistic cloud management approaches that fuse domain specific solutions from areas such as worklo ad prediction, autonomic system management, and empirical analysis.