Multi-dimensional Resource Integrated Scheduling in a Shared Data Center

Abstract

Resource scheduling is crucial to data centers. However, most existing resource scheduling algorithms focus only on one-dimensional resource models, ignoring the fact that multiple resources (e.g. CPU, memory, storage, and network bandwidth) are consumed simultaneously. Competition for multiple resources has become increasingly severe as cloud computing allows uncoordinated and heterogeneous users to share a data center. In this paper, we map such a resource scheduling problem to a bounded multi-dimensional scheduling problem (B-MDKP), taking into account the requirement dependency among multi-dimensional resources. Due to the NP hardness of B-MDKP, we present Multi-dimensional Resource Integrated Scheduling (MRIS), a novel heuristic algorithm to obtain the approximate optimal solution. To demonstrate the advantage of our scheduling scheme, we have implemented MRIS in Haizea, a resource management architecture and made a detailed comparison with existing studies. Our evaluation results show that MRIS achieves high efficiency and high performance for a diverse set of workloads.