Optimal application allocation on multiple public clouds.

Cloud computing customers currently host all of their application components at a single cloud provider. Single-provider hosting eases maintenance tasks, but reduces resilience to failures. Recent research (Li et al., 2010) also shows that providers’ offers differ greatly in performance and price, and no single provider is the best in all service categories. In this paper we investigate the benefits of allocating components of a distributed application on multiple public clouds (multi-cloud). We propose a resource allocation algorithm that minimizes the overall cloud operation cost, while satisfying required service-level agreements (SLAs). In spite of the additional delays for inter-cloud communication and the additional costs for inter-cloud data transfer, our simulation study, using real cloud performance and cost data, demonstrates that multi-cloud allocation outperforms single-cloud allocations in a variety of realistic scenarios.