Online Work Distribution to Clouds

Abstract

The Cloud supports diverse workloads and simple schemes are needed to allocate jobs with satisfactory QoS and low overhead. This paper presents a further study on the potential of an online work distribution approach in adaptively distributing workloads under variable load conditions for optimizing the two contradictory criteria: reducing the energy consumption per job while maintaining the best possible job response time. For cloud systems spanning multiple geographical regions, this paper describes a smart distributed system which deploys a Task Allocation Platform (TAP) in each cloud and takes decisions to allocate tasks dynamically to the service that offers the best overall Quality of Service. Experiments are conducted in dynamic and heterogeneous environments at the global intercontinental level, both to collect data for decision making and to illustrate the effectiveness of our approach.