Due to rapid growth of content-based cloud services (e.g., video streaming), energy usage of cloud infrastructures, consisting of distributed Data Centers (DCs) interconnected by a highbandwidth long-distance network, keeps increasing. This energy usage could further increase due to requested content redundancy to ensure content-service resiliency. Typical content redundancy schemes are based on Content Replication (CR), i.e., each content is replicated in at least one secondary DC location,

more »

... chable in case of failure affecting the primary location, which induces at least 100% increase in storage energy consumption. To reduce storage energy overhead of CR, we investigate a new redundancy scheme, called Content Fragmentation (CF). CF exploits Reed-Solomon erasure code to fragment and encode content into blocks with less storage overhead (thus less storage energy usage) to guarantee content-service resiliency. But CF requires additional energy for content reconstruction and data transport in the core network. To determine which scheme is more energy efficient, we formulate, for both, the contentplacement problem using Mixed Integer Liner Program (MILP), with the objective to minimize energy consumption. Also, due to MILP's poor scalability, we propose a Meta-heuristic Content Placement and Routing Assignment algorithm (M-CPRA) for more efficient solutions. We observe the impact on energy usage of three main metrics, i.e., number of content requests (popularity), resiliency, and latency. Results from a realistic case study suggest that CF always consumes less energy than CR given the same resiliency requirement, and CF is particularly energy-efficient when stored contents are less popular and latency-tolerant.