Fitted Fat-Tree for Localized Traffic in Data Center Networks
Jun Duan, Yuanyuan Yang · 2018
In this paper, we propose a data center network architecture, named fitted fat-tree. Fitted fat- tree provisions unique flexibility in the sense that its topology and performance can be tailored with great freedom. This flexibility allows the architecture to closely follow the demands of the traffic. Given that traffic localization is ubiquitous in today's data centers, the fitted fat-tree can find its wide applications because it can be configured to adapt to the localized traffic load. We first propose a simple and efficient way to profile the localized traffic in the data centers, and then fit the architecture into the profile.We design the topology of the fitted fat-tree, propose its addressing and routing schemes, analyze its cost, and evaluate its performance. Because of its fitted nature, the architecture delivers the same performance as the original fat-tree, but incurs significantly reduced cost, because it uses the network resource much more efficiently.