Fair multi-node multi-resource allocation and task scheduling in datacenter

Abstract

We consider the fair and efficient resource allocation in an environment of multiple resource types and multiple nodes in a datacenter. The heterogeneity of users' resource demands and nodes' resource capacities make it difficult in evaluating the fairness and efficiency of resource allocation and task scheduling. Previous works only considered the fairness measures in the case of a single node. We first propose a novel fairness measure for multiple node resources. We then formulate the resource allocation problem as a convex optimization problem and develop a subgradient algorithm. Our experiments show that the algorithm achieves good efficiency and fairness.