Online scheduling of jobs with favorite machines

Abstract

A long distance link may cause high data latency in cloud computing systems, and thus a computing task may be processed faster by nearby servers than distant servers. To address this class of job (task) scheduling problems, we propose the favorite machine model. Specifically, we are interested in the online version where jobs arrive one by one and must be allocated irrevocably upon each arrival without knowing the future jobs. The objective is to design efficient online algorithms for allocating jobs in order to minimize the makespan.Theoretical performance guarantees are presented for the Greedy algorithm and the Assign-U algorithm, where the latter is shown to be the best-possible online algorithm for this problem. Our theoretical results generalize the results for several classical problems, e.g. the unrelated machines and the identical machines. We also study a restriction of the model, called the symmetric favorite machine model. A 2.675-competitive algorithm is developed and proved to be the best-possible algorithm for the two machines case. Moreover, computational results show that the algorithms perform quite well for random instances, and reveal some insights for choosing algorithms for practical applications.