Performance evaluation of TCP over optical burst-switched data center network

Abstract

In this paper, we consider the performance of the transmission control protocol (TCP) when used in data center networks featuring optical burst switching (OBS). We evaluate the performance using network-level simulation. The bandwidth-delay product in wide-area OBS networks is high enough to require the use of one-way reservation to achieve good throughput. The resulting burst loss will be mistakenly interpreted by the TCP layer as congestion in the network, leading to serious degradation of the TCP performance. The reduced bandwidth-delay product in data center networks allows the use of two-way reservation that results in zero burst loss. We show that this results in improved performance when TCP is used. The modelled switch architecture features fast optical switches in a single hop topology with a centralized optical control plane. We apply different workloads with various burst assembly parameters to evaluate the performance of TCP. Our results show that, with suitable choices for the OBS system parameters, low latency and high throughput performance can be achieved when using TCP.