Performance study of TCP flows with QoS-supported OpenFlow in data center networks

Abstract

Providing QoS guarantees is an important task in Cloud data centers. Applications running on data centers can achieve predictable performance only if the network resource - bandwidth - is also guaranteed along with the computational and storage resources. While QoS guaranteeing solutions in network has been a subject of intense research in the past decades, in this paper, we focus on studying the capability of software-defined networking architecture in providing QoS guarantees in data centers. In particular, we experiment with the most well-known and standardized OpenFlow, to analyze the goodness of the QoS APIs supported in the recent version (OpenFlow v1.3) in providing rate-guarantees to TCP flows in a data center. We experiment with the commonly implemented TCP congestion control algorithms, namely CUBIC and New Reno. Our study demonstrates that, with rate-limiters, the TCP flows that go through an OpenFlow-enabled switch can experience drops to batches of packets, forcing them to switch to the slow-start phase periodically, thereby leading to inefficient bandwidth utilization. This behaviour can adversely affect the performance of applications generating the flows. Hence, we need to develop better mechanisms that limit flow-rates while allowing the flows to operate in their congestion-avoidance phase.