Abstract
As the evolution of distributed computing systems, the need of high performance and large-scale computing is getting increase. Cloud applications or users may demand a storage system with security, availability, performance, and reliability. A variety of public cloud storage providers and private storage systems tried to meet the requirements by applying different approaches. However, no one of cloud storage providers is able to fulfil every requirement of expectations at the same time because of the CAP theorem. In addition, cloud storage providers usually offer different APIs for the access so as to lead users face an issue about vender lock-in. Leveraging heterogeneous storage resources in federated cloud storages is a prospective manner to solve these issues. In this paper, we focus on proposing a federated cloud storage system with a uniform interface. Based on a prioritized brokerage model, a resource brokerage is further presented to benefit the matchmaking with the considerations of user requirements, file classifications, and storage characteristics. We evaluate the system performance with real traces and workloads on 31 nodes. Experimental results show that our approach improve 35%~125% performance gains.