A Scalable Fair Heterogeneous Resource Allocation Scheme in Distributed Systems

Fair and efficient multiple resource allocation is often considered in a data center with a large scale of heterogeneous nodes and requests. Previous work proposed to maximize the minimum resource share received by any user and developed a fast Best-Fit algorithm to allocate jobs to servers. However, Best-Fit algorithm does not always make the optimal assignment. We first apply the subgradient technique to the max-min fairness allocation problem, and further develop two fast online algorithms, the heuristic algorithm and the randomized algorithm, respectively. The latter is completely decentralized and with low complexity. Experiments show that the two algorithms achieve better fairness compared with Best-Fit algorithm.