Hierarchical Approach for Efficient Workload Management in Geo-distributed Data Centers

Geographically distributed data centers (DCs) offer promising business opportunities to both big companies that own several sites and multi-owner inter-cloud infrastructures. In these scenarios, workload management is a particularly challenging task, since the autonomy of single DCs should be preserved while global objectives, such as cost reduction and load balance, should be achieved. In this paper, a hierarchical approach for workload management in geographically distributed DCs is presented. The proposed solution is composed of two algorithms devoted to workload assignment and migration. Both algorithms are based on the computation of a simple function that represents the cost of running some workload in the different sites of the distributed DC. The framework requires a very limited exchange of state information among the sites and preserves the autonomy of single DCs and, at the same time, allows for an integrated management of heterogeneous platforms. Performance is analyzed for a specific infrastructure composed of four DCs, with two goals: 1) load balance and 2) energy cost reduction. Results show that the proposed approach smoothly adapts the workload distribution to variations of energy cost and load, while achieving the desired combination of management objectives.