StorNIR, a Multi-Objective Replica Placement Strategy for Cloud Federations

Federation of clouds makes it possible to transparently extend the resources of Cloud Service Providers (CSPs). For storage services several metrics need to be considered to satisfy customers QoS, that is storage performance, network latency and data availability. Data replication is a key strategy to optimize such metrics. For a CSP, member of a Federation, an effective placement of customers data object replicas is crucial to satisfy QoS demands. In this paper, we modeled the replica placement problem as a multi-objective optimization problem (MOOP) taking into account the local storage classes, other federation CSPs (external) storage services, and customers requirements. To solve this problem, we propose StorNIR a cost-efficient data object Storing scheme based on NSGAII upgraded with Injection and Reparation operators. StorNIR is a matheuristic that consists in hybridizing an exact method with NSGAII meta-heuristic. A repair operator was designed to make the solutions feasible with regards to the system constraints (storage volume, IOPs, etc). StorNIR performed better than both NSGAII meta-heuristic and the exact method in terms of quality of solutions and scalability. The repair function improves the NSGAII meta-heuristic up to 7 times with 7.4% more extra time execution. On average, StorNIR enhances by 17 times the quality of the initial solutions calculated by CPLEX in terms of Hypervolume. In addition, the designed matheuristic approach can be generalized to other meta-heuristics than NSGAII such as MOPSO meta-heuristic.