Wireless Ad Hoc and Sensor Networks

110 Wireless Ad Hoc and Sensor Networksmarking, rather than dropping packets, as indication of congestion at endnodes (Floyd 2001). However, many of the problems (Items 1,2,4, and 5in the previous list) still remain.Congestion control mechanisms in today’s Internet already representone of the largest deployed artificial feedback systems; a protocol shouldbe implemented, in a decentralized way — using sources and links, andwhich satisfies some basic objectives, such as high network utilization inequilibrium and local stability for arbitrary delays, capacities, and routing(Low et al. 2002). Under the assumption that queuing delays will eventuallybecome small, relative to propagation delays, stability results for afluid-flow model of end-to-end Internet congestion control is derived inJohari and Tan (2001). The choice of congestion control mechanism isconstrained to adjust the sending rate on a timescale fixed by the propagationdelay. The end users are naturally placed to implement a controlscheme that acts on the timescale of the end-to-end propagation delay,and thus, one might hope for a stable, distributed end-to-end congestioncontrol algorithm.Another approach, outlined in Floyd and Fall (1999), is for routers tosupport the continued use of end-to-end congestion control as the primarymechanism for best-effort traffic to share scarce bandwidth, and to deployincentives for its continued use. These incentives would be in the form ofrouter mechanisms to restrict the bandwidth of best-effort flows, using adisproportionate share of the bandwidth, in times of congestion. Thesemechanisms would give a concrete incentive to end users, applicationdevelopers, and protocol designers to use end-to-end congestion controlfor best-effort traffic. More recently, there is a surge in the developmentof novel end-to-end congestion control schemes (Jagannathan and Talluri2002, Chandrayana et al. 2002, Bansal and Balakrishnan 2001, Borri andMerani 2004, Kuzmanovic and Knightly 2004, Sastry and Lam 2005), suchas binomial congestion control (Bansal and Balakrishnan 2001), TCPfriendly schemes (Borri and Merani 2004), and CYRF (Sastry and Lam2005), to overcome the limitations of TCP-based networking protocols.However, it was shown through simulation studies in Chandrayana et al.(2002) that AIMD-based schemes still outperform TCP-compliant binomialschemes for congestion control in a wide range of networking environments.In contrast to ATM networks where hop-by-hop typecongestion control is utilized (Section 3.3), end-to-end congestion controlis discussed for the Internet.This chapter presents a new rate-based end-to-end control scheme fromPeng et al. 2006 to overcome some of the problems of the existing Internetcongestion (Bansal and Balakrishnan 2001, Borri and Merani 2004). Thenovelty of the scheme is its distributed network modeling based on nonlinearsystem theory, using which, the performance and stability of thescheme can be mathematically analyzed and established. The proposed