ECN-based congestion probability prediction over hybrid wired-wireless networks

Abstract

Network congestions and wireless link errors are both potential reasons for packet losses in hybrid wired-wireless networks. Their coexistence necessitates the design of new mechanisms that can differentiate network states more precisely. In this paper, a method is proposed to distinguish between the two reasons of packet losses in hybrid wired-wireless networks (i.e., congestions and wireless link errors) and calculate the probability of network congestion in the former case. This method combines the information contained in CE bits of a sequence of ECN-enabled acknowledge packets to calculate the probability of network congestion; thus it is more accurate than methods using CE bits of a single acknowledge packet. Analysis on the effectiveness of the proposed method in calculating congestion probability is performed via simulations. The ability to differentiate precise network states helps a TCP source response to ECN feedback more adaptively. We discuss how to enhance existing TCP variants with the proposed method. Simulation results show that the enhanced TCP variants can effectively improve network performance. © 2014 Jin Ye et al.