Nonlinear Control Analysis of Quantized Congestion Notification in Data Center Networks

Abstract

In data center networks, quantized congestion notification (QCN) has been ratified as the standard layer 2 congestion control mechanism. Since the congestion control algorithm of QCN conducts nonlinear switching between the rate increasing and decreasing stages, it is very difficult to analyze QCN theoretically. Previous works derive the stability condition of QCN using tools from linear control theory. However, the tools cannot reveal the reason why the queue at a congested link oscillates persistently when QCN is enabled. In this paper, we use the describing function method from nonlinear control theory to analyze queue oscillation. We show that unsuitable network configurations like more flows, a higher link capacity and a larger link delay will cause more severe oscillation of the queue. In particular, the impacts of the QCN parameters, such as the sampling probability and the constant in the active increase phase in QCN, on the oscillation magnitude of the queue are also analyzed. The simulation results validate the conclusion from our theoretical analysis well.

Keywords

• Quantized congestion notification
• data center networks
• nonlinear analysis
• describing function
• limit circle