Online Non-preemptive Multi-Resource Scheduling for Weighted Completion Time on Multiple Machines

Donney Fan, Ben Liang

Abstract: Jobs in computing environments have diverse and heterogeneous resource requirements. This paper presents a study of online, non-preemptive scheduling algorithms for multiple identical machines under the average weighted completion time objective. The key challenge addressed is resource allocation to jobs with non-uniform demands across multiple resource types, such as CPU, memory, and storage. We propose an online algorithm, termed Multi-Resource Interval Scheduling (MRIS) that achieves a competitive ratio of $8R(1 + \epsilon)$ for the average weighted completion time, where R is the number of resource types. To the best of the authors’ knowledge, this is the first theoretical competitive analysis under the considered system. We further show that the well-known priority queue algorithms can have arbitrarily bad competitive ratios in this setting. In numerical experiments using production workload traces from Microsoft Azure, the proposed algorithm is shown to significantly outperform priority queue algorithms and other state-of-the-art schedulers.