Performance Study of Three Different Types of Data Center Network Architectures, NovaCube, BCube, and FatTree Using NS-3 Simulator

Abstract

In Data Centers, the Data Center Network (DCN) is an important component, which is an interconnection of servers and switches with high-speed communication links. As demands of cloud-based services are increasing, the scale of the DCN has been increasing and cost and power consumption have became one of the major concerns. Existing fundamental DCN architectures such as FatTree, BCube use external switches to interconnect servers, which are cost inefficient and more power-consuming. In order to deal with these problems, a novel torus based DCN architecture named NovaCube and Probabilistic Oblivious Routing Algorithm PORA for NovaCube was recently proposed. However, effectiveness of this proposed architecture of NovaCube and PORA algorithm had been given only theoretically. In order to take better business decisions, improve the architecture, or to introduce an entirely new architecture, in this paper, we presented performance studies of three different types of DCN Architectures named FatTree, BCube, and NovaCube. NS-3 platform had been used to implement these architectures. Performances of these DCN architectures are measured in terms of average throughput, average latency, cost, power, and scalability.