The Research Of Congestion Control Algorithm Based On Optimization Flow Control

With the rapid development of computer network, the Internet users increase sharply, the network congestion also changes more and more seriously. Since Jacobson proposed the congestion control concept in 1988, the congestion control has aroused people's enormous interest. Each kind of congestion control theory as well as the algorithm also one after another proposed. Congestion control now is hot topic of Internet research. The congestion control research as a result of its huge complexity, the existing congestion control algorithm has not satisfied formerly based on the subjective analysis method, but is in virtual of the strict mathematics proof, the method of non-linear dynamics, the control and the optimization theoretical analysis complex changeable network system, to design new congestion control algorithm adapting for the network.The rationale of this paper is the non-linear optimization flow control, the optimization flow control theory is from the microeconomics, the game theory and the non-linear optimization theory, Kelly presented the foundational paper of optimized flow control frame in1998, Low proposed the integrated theory of optimized flow control based on this in December1999. Its key idea is: the objective of the optimized flow control is to maximize aggregate source utility over their transmission rate. It views network links and sources as processor of a distributed computation system, Low et al solve the dual problem of optimization problem using the asynchronous gradient projection algorithm, an essential characteristic is this problem may decompose some simple sub-problem, given the price of dual problem, individual source can solve its maximum benefit of primal problem separately without the need to coordinate with other sources. In this system sources select transmission rates that maximize their own benefits, and network links adjust bandwidth prices to coordinate the sources' decisions. Low et al based on the theory of optimization flow control, proposed the duality model of TCP/AQM, the basic idea is to regard source rates as primal variables and congestion measures as dual variables, source adjusts its transmission rate by congestion measures, but various sources transmission rate influence congestion measures in turn, thus constitutes a closed loop congestion control system. This model regards the existing TCP congestion control and the AQM algorithm as a distributed algorithm which solves optimal rate assignment of suitable utility function, thus may theoretically analyze performance of network in the equilibrium, such as throughput, loss rate, time delay and queue length.This paper comprehensively analyzes and discusses optimization flow control theory, combines source algorithm and link algorithm effectively based on duality mode of TCP/AQM, and proposes two congestion control algorithms based on the optimized flow control. Its main work includes:First, this paper comprehensively analyzes and studies the theory of optimization flow control. This is the rationale of congestion control algorithm in the paper. In chapter 2 we study the theory of flow control. Firstly, this chapter introduces the theory frame of optimization flow control, presents primal-dual problem of optimization flow control and gradient algorithm solving dual problem. Secondly, this chapter presents duality mode of TCP/AQM based on optimization flow control and studies its mathematics model and main ideology carefully. Finally, we has discussed the REM algorithm, we have analyzed the REM characteristic and its parameters establishment.Second, this paper proposes a new congestion control algorithm based on optimization flow control: MReno. Aiming at periodic oscillation and inefficient use of the available bandwidth of TCP Reno and unifying optimization flow control, we propose a new congestion control algorithm: Marking Reno(MReno). The NS-2 simulation experiments prove that the MReno algorithm has a better stability, bandwidth use and fairness than the Reno algorithm. This is our expectant result which conforms to the idea of optimization flow control. But, as well as the Reno algorithm, the stability of MReno algorithm will decrease with the bandwidth increasing. This is also a problem which we need to solve in the future.Third, this paper proposes a new congestion control algorithm based on the optimization flow control: PVegas. As an alternative to TCP Reno algorithm, unlike Reno(or its Variants such as NewReno and SACK), that uses packet loss as a measure of congestion, Vegas use queue delay as a measure of congestion. TCP Vegas has obvious advantage and shortcoming. TCP Vegas has high throughput when only TCP Vegas connections exist in the network and has no loss in the balance. Moreover, TCP Vegas has good fairness to connections of long delay. Certainly, TCP Vegas has its own shortcoming. While competing for limited bandwidth and non-TCP Vegas connection, TCP Vegas will be extremely disadvantageous condition unless the buffer of router is very small. Furthermore, TCP Vegas may suffer persistent congestion. Aiming at the shortcoming of TCP Vegas, we propose a new congestion control algorithm based on the optimization flow control: Price Vegas(PVegas). The NS-2 simulation experiments prove that PVegas will make source obtain bigger utility and solve the persistent congestion effectively. PVegas has better fairness, stability and lower packet loss than TCPVegas.