A new high-performance TCP friendly congestion control over wireless networks

Abstract

In recent years, new delay-based congestion control mechanisms like Caia-Hamilton Delay (CHD) or Caia Delay-Gradient (CDG) are becoming popular due to their effectiveness even over wireless links. CHD and CDG are designed to behave friendly with conventional TCP NewReno flows. However, due to their delay-based window update mechanisms, CHD or CDG performance is often vulnerable to the co-existing aggressive TCP NewReno flows. In addition, in case of CHD, it is practically very difficult to choose the optimum values of tuning parameters with a view to ensuring the expected performance. In this paper, we therefore propose a new mechanism to overcome the above issues. We name it as Wireless Friendly Congestion Control (WFCC). Our objective is to devise a congestion control scheme that is (i) friendly with TCP NewReno flows over wireless links when deployed together, (ii) free from delicate operational parameters and (iii) robust against link errors under a wide range of network buffer space. We conduct a thorough evaluation of our proposed approach by simulation as well as emulation. Results show that WFCC (i) can lead up to 250% improvement in performance compared to TCP NewReno schemes and (ii) are friendly with TCP NewReno flows over wireless links.

Introduction

In recent years, mobile and wireless devices are being incorporated to the Internet at a rapid pace. This initiates the need for optimizing major applications to perform satisfactorily over wireless link. Wireless links are generally prone to a higher link error rate than their wired counterparts. Since conventional TCP protocols like TCP NewReno are designed to be used in wired networks with low link error rates, their original designs do not take link errors into account. Hence, they assume that segment losses occur solely due to congestion in networks (Jacobson, 1988). Whenever a segment loss occurs, conventional TCP infers congestion and the TCP sender reduces its sending rate as a remedial measure. Therefore, when conventional TCPs are used over wireless networks, they interpret all the losses to be originating from congestion and frequently lower their transmission rates unnecessarily. As such, deployment of conventional TCP over wireless links results in a decreased throughput (Balakrishnan et al., 1997).

Various end-to-end congestion control mechanisms have so far been proposed to handle this problem. Most of these mechanisms are reactive. That is, whenever a segment loss occurs, they attempt to figure out whether the actual reason for the loss was network congestion or wireless link error. We therefore mention them as ‘loss based’ schemes in this paper. In addition, a comparatively new ‘proactive’ approach for TCP congestion control has been flourishing in recent years. They aim at decoupling TCP performance from the network buffer space by using delay in the network as the parameter. We call them ‘delay-based’ approaches. Among these delay-based approaches, mechanisms like Caia-Hamilton Delay (CHD, Hayes and Armitage, 2010) and Caia Delay-Gradient (CDG, Hayes and Armitage, 2011) offer effective solutions for wireless link loss.

CHD and CDG are designed to behave friendly with conventional TCP NewReno flows. However, due to their delay-based window update mechanisms, CHD or CDG performance is often vulnerable to the co-existing aggressive TCP NewReno flows that ultimately occupy most of the bottleneck link bandwidth. In addition, delay based mechanisms are highly sensitive to the tuning of corresponding parameters. For example, in case of CHD, it is not possible to determine the most accurate value of the queuing threshold intuitively.

Previously, we proposed very TCP friendly congestion controls over wireless links, named as Utilization-based Congestion Control (UCC, Utsumi and Zabir, 2012). UCC provides a solution based on the utilization at the bottleneck link. We estimate the utilization on the bottleneck link using queueing theory and apply this for end-to-end congestion control.

In this paper, we propose a novel congestion control mechanism to overcome the above issues. We name this as Wireless Friendly Congestion Control (WFCC). Our proposed mechanism, WFCC, is based on the queueing delay at the bottleneck link and is (i) friendly with TCP NewReno flows over wireless links when deployed together, (ii) free from delicate operational parameters and (iii) robust against link errors under a wide range of network buffer space.

We conduct a thorough evaluation of our proposed approach by emulation. Results show that (i) WFCC can yield a performance improvement of 170% or more compared to TCP NewReno schemes and a performance improvement of 93% or more compared to UCC schemes over wireless links and (ii) it is friendly with TCP NewReno flows over wireless links.

The rest of this paper is organized as follows. Section 2 summarizes the most recent relevant research works. In Section 3 we present how one way delay can be utilized for deriving our other proposed mechanism, Wireless Friendly Congestion Control (WFCC). In Section 4, we discuss in detail how we have evaluated WFCC through emulation. Finally we conclude in Section 5.