The Case for Fine-Grained Traffic Engineering in Data Centers

Abstract

Data center traffic characteristics are not well understood. In particular, it is not clear how the prevalent traffic patterns may impact candidate mechanisms for managing traffic inside the data centers. In this paper, we conduct a measurement study of network-level traffic patterns inside data centers. Based on our empirical insights, we design a traffic generator for creating representative workloads for traffic between TOR switches in a data center. We use this generator to evaluate several traffic engineering techniques and data center network architectures, and analyze their short comings. Our findings highlight the need for fine-grained traffic engineering (TE) mechanisms. We design and implement such an approach using OpenFlow and show how it can significantly improve data center TE.