Policy-Driven Service Placement Optimization in Federated Cloud

Efficient provisioning of elastic services constitutes a significant management challenge for cloud computing providers. We consider a federated cloud paradigm, where one cloud can subcontract workloads to partnering clouds to meet peaks in demand without costly over-provisioning. We propose a model for service placement in federated clouds to maximize profit while protecting Quality of Service (QoS) as specified in the Service Level Agreements (SLA) of the workloads. Our contributions include an Integer Linear Program (ILP) formulation of the generalized federated placement problem and application of this problem to load balancing and consolidation within a cloud, as well as for cost minimization for remote placement in partnering clouds. We also provide a 2-approximation algorithm based on a greedy rounding of a Linear Program (LP) relaxation of the problem. We implement our proposed approach in the context of the RESERVOIR architecture.