Design of On-Line Tuned Controller for Congestion Avoidance in Computer Networks

Active Queue Management (AQM) applies a suitable control policy upon detecting congestion in networks. In this paper, a Proportional-Integral (PI) controller based on the optimization algorithm of MATLAB/Nonlinear Control Design Blockset (NCD) which is adapted as On-line tuning for controller parameters is applied to AQM for the objective of congestion avoidance and control in middle nodes. To present the methodology, a Proportional Integral (PI) controller is verified with the TCP/AQM model as an Active Queue Management (AQM) in internet routers The analytical results for linearized TCP/AQM model are presented in MATLAB/Simulink. From the obtained results, a faster response time as well as the regulation of the output to a constant value by the PI controller is clearly observed and it is noted that the PI controller based NCD provides good tracking performance for congestion avoidance in computer networks


TCP/AQM MODEL
AQM has been extensively analyzed using control-theoretical methods.Controltheoretical approaches lead to stable, effective, and robust congestion control operation.In [5], the non-linear dynamic model for multiple TCP flows control has been developed based on fluid-flow theory to model the interactions of a set of TCP flows and AQM routers in computer networks which consist of a system of nonlinear differential equations.For the control theoretical analysis, it was approximated as a linearized constant model by small signal linearization about an operating point( W 0 , q 0 , p 0 ), see [6,12] for linearization details, which leads to the following :     From Figure ( 7) it is shown that the system without controller is unable to track the queue length around the queue length to the desired level, where the system goes into a sustained oscillation with high congestion exceeding the maximum buffer size.In order to eliminate this sustained oscillation and get better tracking performance a classical control (PI-controller) is applied.As it is shown in Figure ( 8) below.The system response with PI controller has the ability to get the desired performance, where the desired queue length is achieved within the buffer size of the router.The PI controller parameters are selected by trial and error method.Thus the PI coefficients are K p = ٠.٠٠٠٠٢٥ and K i = ٠ .٠٠٠٠٠٨ respectively.Although the PI controller shows good performance the system response is slow, so to overcome this drawback and to get the best parameters to controller and for enhancing the system response the NCD is used as on-line tuning method to get best Parameters for PI controller to speed up the system response.

CONCLUSIONS
From the design and the simulation results, it can be concluded that: 1-The applied PI controller based NCD method can deal with congestion problem with a good tracking performance as shown from the parameters obtained in Table (1) about the desired queue size with high link utilization and faster system response observed.2-By using the NCD to tune PI parameters the system response is improved as shown in table () which prove the efficiency of NCD method.3-NCD is adapted for real-time executions and it can tune any controller without the mathematical model knowledge of the system it is controlling.

INTRODUCTION ctive queue management
refers to the practice of manipulating the queue at an outbound interface in a router to bias the performance of flows that transit the router.The goals of active queue management are (1) to reduce the average length of queues in routers and thereby decrease the end-to-end delay experienced by packets, and (2) to ensure that network resources are used more efficiently by reducing the packet loss that occurswhen queues overflow[1].Congestion in a computer network is a state in which performance degrades due to the saturation of network resources such as communication links, processor cycles, and memory buffers.Network congestion has well recognized as a resource-sharing problem.When too many packets are contending for the same link, the queue overflows and packets have to be dropped.When such drops become common events, the network is said to be congested.A brief description of these researches is submitted in the following paragraphs.Jacobson and Karels [2] proposed the end-to-end congestion control is algorithms which forms the basic for the TCP congestion control.Its content that a TCP sender keeps a sending window (packets) rate according to the rate of dropped packets when a buffer becomes full in the router queue.Floyd and Jacobson [3] presented the RED.Its mechanism is that packets are randomly dropped before the buffer of queue overflows.And, Braden et al. [4] proposed the enhanced end-to-end congestion control for AQM.Misra et al.[5] developed a methodology to model and obtain numerically expected transient behavior of networks with AQM routers supporting TCP flows.The solution methodology scales well to a large number of flows.This modeling/solution methodology has a great potential in analyzing and understanding various network congestion control algorithms.Hollot et al. [6] used linearization to analyze a previously developed Non-linear model of the TCP/AQM.They linearized the model by using small-signal linearization about an operating point to gain insight for the purpose of feedback control to analyze a combined TCP and AQM model from a control theoretic standpoint and ability to present design guidelines for choosing parameters that lead to stable operation of the linear feedback control system.Hollot et al. [7] proposed PI controller based on the linear control theory.The main contribution is to convert the congestion control Algorithm into the controller design problem within the framework of control theory in AQM system by studying a previously developed linearized model of TCP and AQM.The controller showed better theoretical properties than the well known RED controller.Yuanwei Jing et al. [8] developed a new AQM algorithm based on Fuzzy Sliding Model Controller for the objective of congestion avoidance and control in middle nodes.The proposed controller combines the excellent characteristics of linear sliding model controller (LSMC) and the terminal sliding model controller (TSMC).The LSMC is used to speed up the error convergence when the error is greater the TSMC is adopted to guarantee the error convergence to zero in a finite time when the error is around the zero.Yann et al. [9] Proposed an AQM based on the Lyapunov theory for time delay systems.With the help of Lyapunov-Krasovskii functional and using a state space representation of a linearized fluid model of TCP.Note that the proposed methods have been extended to the robust case where the delay in the loop is unknown.Several researches have been done in the field of congestion avoidance in traffic of computer networks.Kang et al. [10] proposed the LQ-Servo controller for AQM routers.The proposed controller structure is made by taking a traditional servo mechanism based on Linear Quadratic (LQ) approach.The proposed LQ-Servo controller can deal with a good tracking performance comparing with PI controller.Kang et al. [11] developed the LQ-Servo controller based on loop shaping method for TCP/AQM router in order to meet such frequency domain design specifications as good disturbance rejection.The simulation results show that the proposed controller is more effective in getting the good tracking responses than PI controller for the varying reference queue size in AQM routers.
the time derivative of q(t) , and W :Expected TCP window size (packets) q : Expected queue length (packets) R 0 :Round-trip time (seconds) C :Link capacity (packets/second) N : Load factor (number of TCP sessions) ٥٧٦ p :Probability of packet mark/drop t : Time The expected queue length q and the expected TCP window size W are positive value and bounded quantities.And also, the probability of packet (mark /drop) p takes value only in [0 ,1] .Taking the Laplace transform of equation (1) and rearranging the following transfer functions are obtained: block diagram of linearized AQM control system is shown in Fig.1.In this diagram Ptcp (s) denotes the transfer function from loss probability δp(t) to window size δW(t) , Pqueue (s) denotes the transfer function from δW(t) to queue length δq(t) , and C(s) denotes the transfer function of controller.

Figure
Figure (4): block diagram of Engine Model with PI Controller and NCD blockset.

Figure
Figure (9): System Response with PI Controller based on NCD.From Figure (9) above it is shown that the PI controller based NCD speeds up the response comparing with PI controller, where PI controller parameters are get from NCD as follows K p = ١.٢٨٩٣e-005, K i = ٩.٥٠٣٣e-006.