Utility-Driven Bandwidth Allocation in Data Center Networks

Abstract

It has been realized that network performance is one of the most important metrics for evaluating the performance of applications deployed in cloud data centers. Numerous bandwidth allocation schemes have been proposed to maximize the utilization of data center networks as well as to improve the performance of applications. However, those works only focus on allocating link bandwidth shared among elastic applications; the path delay experienced by inelastic traffic such as delay-sensitive traffic has been neglected. In this paper, we investigate the problem of maximizing application utility in data center networks considering both the throughput and the delay influence. The utility of an application is the benefit brought by bandwidth increases minus the expenditure charged by congestion delay growth. We then formulate the utility-driven bandwidth allocation problem as a convex optimization with the objective of maximizing overall utility across all applications. Standard interior point algorithm is applied to derive the optimal solution. We show the outstanding performance of our solution through extensive simulations with several realistic data center network (DCN) topologies.