Hybrid Storage Throughput Allocation Among Multiple Clients in Heterogeneous Data Center

The SSD is adopted to improve the IO performanceof the storage system in the data center, the throughput allocation for clients is a challenging problem. We need to find a throughput allocation method, which can determine the throughput allocations of clients on each server, while both providing the fair allocations for clients and maximizing the utilization of system throughput resource in the entire data center. In this paper, we propose an unified allocation framework(UAF), UAF jointly allocates throughput resource across allheterogeneous storage servers, can achieve the fair throughput allocations for all clients and maximize the utilization of system throughput resource in the entire data center. We can prove that UAF satisfies fairness properties of Envy Freedom, Sharing Incentive and Pareto Optimality. We evaluated the performance of UAF using both simulation and implementation in a cluster consisting of multiple heterogeneous storage severs. The experiment results show our method can determine the throughput allocations of clients in each server, and provide both the efficient and fair allocation in the entire data center.