The purpose of this paper is to design congestion controller for TCP/AQM (transmission control protocol/active queue management) networks using model following control; the equilibrium of a class of TCP/AQM networks with time-delay is investigated, and the effect of communication time-delay on the stability is addressed. The features of this design method are bounded property of the internal states of the control system being given and the utility of this control. Such design exhibits important attributes including fast convergence with high accuracy to a desired queue length. Simulation results show that the time-delay nonlinear behavior of the system can be controlled by this method.
1. Introduction
In recent years, with the rapid growth of throughput-demanding applications, congestion control has emerged as a major issue in computer and communication network design [1]. So many researchers are seeking some methods to effectively control congestion. TCP congestion control mechanism is used to prevent congestion collapse.
AQM schemes have been proposed to complement the TCP network congestion control [2]. Several mathematical models are developed by some researchers [3–5] and a variety of control theory-based AQM schemes are proposed based on these models. The simulated approaches contain a wide range of variations in network topologies, topological parameters, load and capacity, and traffic mixtures. The outperformance of the PFC-AQM in comparison with the commonest AQM methods such as the RED (random early detection), PI, and REM (random early marking) emphasizes the proper applicability of PFC as an AQM method [6–9]. Based on the system model for congestion control in transmission control protocol TCP/AQM networks, control theory-based approaches are utilized either to analyze or to design the AQM schemes. Based on the system model, several conventional controllers [10–15] are designed as AQM methods in TCP networks.
The design of some communication systems requires the implementation of time-delays within the system. These time-delays can be accomplished with a variety of optics technologies, which could be readily fabricated and integrated into the communication system without significant impacts on the system design [16–19]. Time-delay is very important for the modeling of networks, occurring both in the control of networks and in the control over networks [20]. In the context of communication networks, the term “congestion control” is generally used to refer to the action of regulating various flows within a network. In recent years, intense research efforts are devoted to the application of the Smith predictor for queue length control of ATM (asynchronous transfer mode) networks [21]. The TCP was designed in the late 1983s by Jacobson, which is a critical part of the internet machinery.
The purpose of this paper is to design a congestion controller based on the model following control system (MFCS) [22, 23] control theory. The features of this design method are that bounded property of the internal state of the system, which is given and confirmed on basis of a numerical example of the network congestion system in which the output signal of the control system asymptotically follows the reference model signal in the case of the existence of disturbances.
The paper is organized as follows. In Section 2, the TCP/AQM network in congestion control model is described. In Section 3, controller design of the network congestion system with time-delay is proposed. In Section 4, bounded analysis of control system internal state is shown. Section 5 is the simulation results. The paper is concluded in Section 6.
2. The TCP/AQM Networks in Congestion Control Model
In this paper, the network in Figure 1 is considered. The network consists of n nodes (sender), 1 node (receiver), and 1 bottleneck router. The bottleneck router sends packets from these senders to the receiver. This network topology denotes 1 server machine to multiple client machines in a computer network. TCP is only the communication protocol in Figure 1. Large-scale networks can be simplified as in Figure 1 in case of designing congestion controllers if only one router is bottleneck in the large-scale computer network.
The network of the congestion control.
In this approach, we overview the dynamical fluid-flow model developed by [24–29] to describe the behavior of TCP/AQM networks. A simplified version of that system model is considered, which ignores the timeout and slow start mechanism of TCP. The model involves the average value of key network variables and is described by the following coupled nonlinear differential equations with time-delay [25]:
(1)W˙(t)=1R(t)-W(t)2W(t-R(t))R(t-R(t))p(t-R(t)),(2)q˙(t)=W(t)R(t)N(t)-D(t),(3)R(t)=q(t)D(t)+Tp,
where W(t) is the congestion window size in packets at time t, q(t) is the queue length at the congested router in packets, R(t) is the RTT (round trip time) which represents the time-delay in TCP dynamics in seconds, D(t) is the link capacity in packets per second, Tp is the propagation time-delay in seconds, N(t) is the number of active TCP connections, and p(t) is the packet mark/drop probability.
These differential equations in the block diagram of Figure 2 are taken from [24] highlighting TCP window-control and queue dynamics.
Block diagram of TCP congestion avoidance mode.
Set up a model for nonlinear TCP networks dynamic model. Let x1(t)=q(t)-q0, x2(t)=x˙1(t), and u(t)=p(t), where q0 is a desired queue length in the router. Assume that the rate of the change for x1(t), N(t), C(t), and R(t) is slower than W(t), q(t); (1) can be expressed in the following form [30]:
(4)x˙(t)=f(x(t))+g(x(t))u(t-R(t)),
where
(5)x(t)=[x1(t)x2(t)]T,f(x(t))=[f1(x(t))f2(x(t))]T,g(x(t))=[g1(x(t))g2(x(t))]T,f1(x(t))=x2(t),f2(x(t))=N(t)R2(t),g1(x(t))=-[(x2(t)+C(t)]+[x2(t-R(t))+C(t)]2N(t).
Therefore, the system is defined by
(6)x˙(t)=f(x(t),u(t-R(t))),(7)y(t)=[01]x(t).
3. Controller Design of the Network Congestion System with Time-Delay
Based on (6), the system in (4) can be rewritten in equivalent form as follows:
(8)x˙(t)=A11x(t)+A12u(t)+f1(x(t),u(t-R(t))),u˙(t)=A21x(t)+A22u(t)+g1(x(t),u(t))+Bu1(t),
where A11∈Rn×n, A12∈Rn×l, A21∈Rl×n, and A22∈Rl×l.
Then, from (8), we can obtain the following system:
(9)x˙1(t)=A1x1(t)+B1u1(t)+f2(x1(t)),y(t)=C1x1(t),
where
(10)x1(t)=[x(t)u(t)]T,f2(x1(t))=[f1(x1(t))g1(x1(t))]T,f1(x1(t))=f(x1(t))-A11x(t)-A12u(t),A1=[A11A12A21A22],B1=[0B]T,C1=[C0].
Let p=d/dt; we have
(11)y(t)=C1[pI-A1]-1B1u1(t)+C1[pI-A1]-1f2(x1(t)).
Then the representations of input-output are given as
(12)D(p)y(t)=N(p)u1(t)+Nf(p)f2(x1(t))=Nrpn+l-2u1(t)+N1(p)u1(t)+C1pn+l-2ddtf2(x1(t))+Nf1(p)f2(x1(t)),
where N(p)/D(p)=C1[pI-A1]-1B1,
(13)Nf(p)=C1adj(pI-A1),N(p)=Nrpn+l-2+N1(p),Nf(p)=C1pn+l-1+Nf1(p).
Now we have
(14)D(p)y(t)=pn+l-2(C∂f(x1(t))∂uT(t)B)u1(t)+N1(p)u1(t)+pn+l-2×{C(∂f(x1(t))∂uT(t)-A12)×(A21x(t)+A22u(t)+g1(x1(t)))+C(∂f(x1(t))∂xT(t)-A11)f(x1(t))}+Nf1(p)f2(x1(t))=pn+l-2(C∂f(x1(t))∂uT(t)B)u1(t)+N1(p)u1(t)+pn+l-2{C(∂f(x1(t))∂uT(t)-A12)×(A21x(t)+A22u(t)+g1(x1(t)))+C(∂f(x1(t))∂xT(t)-A11)f(x1(t))}+C1Γn+l-2pn+l-2f2(x1(t))+Nf2(p)f2(x1(t)),
where Nf1(p)=C1Γn+l-2pn+l-2+Nf2(p).
Let
(15)Γn+l-1=I,Γn+l-2=A1Γn+l-1+αn+l-1I=A1+αn+l-1I=A1-Tr(A1)I,
Where αn+l-1=-Tr(A1).
Now the representations of input-output (12) are given as follows:
(16)D(p)y(t)=pn+l-2(C∂f(x1(t))∂uT(t)B)u1(t)+N1(p)u1(t)+pn+l-2f4(x1(t))+Nf2(p)f2(x1(t)),
where
(17)f4(x1(t))=C{(∂f(x1(t))∂xT(t)-Tr(A1)I)f(x1(t))+(∂f(x1(t))∂uT(t))g1(x1(t))+(∂f(x1(t))∂uT(t)-A12)·(A21x(t)+A22u(t))+(Tr(A1)I-A11)·(A11x(t)+A12u(t))(∂f(x1(t))∂xT(t)-Tr(A1)I)}.
Let Nrvz(t)=(C(∂f(x1(t))/∂uT(t))B)u1(t)+f4(x1(t)).
We have
(18)D(p)y(t)=pn+l-2Nrvz(t)+N1(p)u1(t)+Nf2(p)f2(x1(t)).
The reference model is given as
(19)x˙m(t)=Amxm(t)+Bmrm(t)ym(t)=Cmxm(t).
Choose a stable polynomial T(p) which satisfies the following conditions. (1) The degree of T(p) is ρ≥nd+2n-nm-2-ηi. (2) The coefficient of the maximum degree term of T(p) is the same as D(p).
Consider the following equation:
(20)T(p)Dm(p)=Dd(p)D(p)R(p)+S(p),
where the degree of each polynomial is ∂T(p)=ρ, ∂Dm(p)=nm, ∂Dd(p)=nd, ∂D(p)=n, ∂R(p)=ρ+nm-nd-n, and ∂S(p)≤nd+n-1. T(p), Dm(p), Dd(p), D(p), and R(p) are monic polynomials.
In this paper, we propose a design of model following control system with disturbances. We can prove that all the internal states are bounded and output error
(21)e(t)=y(t)-ym(t)
converges to zero asymptotically. Then the following form is obtained:
(22)T(p)Dm(p)e(t)=Dd(p)R(p){(x1(t))pn+l-2Nrvz(t)+N1(p)u1(t)+Nf2(p)f2(x1(t))}+S(p)y(t)-T(p)Dm(p)ym(t),
where N1(p)=∑k=0n+l-3C1ΓkpkB1=Nf2(p)B1.
The control law (controller) vz(t) can be obtained by making the right-hand side of (22) equal to zero.
Therefore, vz(t) of (22) is obtained from e(t)=0. The model following control system can be realized if the system internal states are bounded.
4. Bounded Analysis of Control System Internal State
Let z1(t)=[x1T(t)ξ1T(t)ξ2T(t)ξ3T(t)]T; then the system is defined by
(24)z˙1(t)=As+ψ(z1(t)),x1(t)=Csz1(t),
where
(25)As=[A1-B1E2C1-B1H1-B1H2-B1H3-G1E2C1F1-G1H1-G1H2-G1H3G2C10F20-G3B1E2C1-G3B1H1-G3B1H2F3-G3B1H3],Cs=[I000],ψ(z1(t))=f2(x1(t))-B1(C∂f(x1(t))∂uT(t)B)-1iiiiiiiiiiiiiii·(W(x1(t))vz(t)+f4(x1(t))),ψ(z2(t))=G3((C∂f(x1(t))∂uT(t)B)-1f4(x1(t)))f2(x1(t))iiiiiiiiiiiiiiiiiiiiiii-B1(C∂f(x1(t))∂uT(t)B)-1f4(x1(t)))iiiiiiiiiiii=G3B1(C∂f(x1(t))∂uT(t)B)-1W(x1(t))vz,C∂f(x1(t))∂uT(t)B=Nr+W(x1(t)).
The characteristic polynomial |pI-As| can be calculated as follows:
(26)|pI-As|=|Q(p)|2Vs(p)T(p)lDm(p)l
with stable polynomials of T(p), Dm(p), |Q(p)|, and Vs(p). Therefore, As is also a stable system matrix.
Now, the system can be rewritten as
(27)z˙2(t)=Asz2(t)+ψ(z1(t)),(28)x1(t)=Csz2(t).
Consider a quadratic Lyapunov function candidate:
(29)V(t)=12z2T(t)Psz2(t),(30)V˙(t)=z2T(t)Ps(Asz2(t)+ψ(z1(t)))=-12z2T(t)Qsz2(t)+z2T(t)Psψ(z1(t))≤z2T(t)Psψ(z1(t))<0,(31)PsAs+AsTPs=-Qs,
where Qs and Ps are symmetric positive definite matrices defined by (27). If As is a stable matrix, we can get a unique Ps from (31) when Qs is given. Therefore, z2(t) is bounded.
5. Simulation Results
This simulation, which proposes static state feedback AQM controller for the time-delay system, verifies the performance.
The network topology is simple and it is shown in Figure 3. But this small-scale network is developed by using four computers, and some experiments are carried out.
The network topology.
This network consists of two senders with four data flows (this means four sessions and N=8), one bottleneck router, and one receiver, which receives data from senders through the bottleneck router (see Table 1).
Sender 1-2 node and receiver node parameters.
Parameter
Value
Application
ftp
Transport layer protocol
TCP
TCP agent SACK bit
False
Packet size
1448 (byte)
Acknowledgment size
40 (byte)
Flow number (TCP session number) (disturbance)
8 (+1) (flow)
In the responses (Figures 4, 5, and 6) of the congestion system with time-delay, the output errors converge to zero. So the effectiveness of this method is verified.
Responses of the congestion system with time-delay (the reference queue size is q-0=1.7 (packets); time-delay is 0.12 (ms)).
Responses of the congestion system with time-delay (the reference queue size is q-0=2.7 (packets), time-delay is 12 (ms), and d(t) is white noises).
Responses of the congestion system with time-delay (the reference queue size is q-0=3.3 (packets), time-delay is 0.1 (ms), and d(t) is white noises).
6. Conclusions
In this paper, a new design method for the congestion controller of the TCP/AQM networks is introduced. The developed approach can theoretically guarantee the system performances, including the disturbance rejection and the implied stability of the closed-loop system. This property is useful for congestion controller design. This paper studies a control system with time-delay using a model following method which is one of the effective means of solving time-delay problems in a control system. The method can efficaciously control time-delay under disturbances and has excellent practicability.
By using this model, the nonlinear input time-delay system, which describes a TCP/AQM network, is transformed into an equivalent nonlinear system, and it is possible to design controllers based on nonlinear control theories. For a congestion control problem, a round packet trip time is not stationary and depends on the actual traffic. Finally, it is pointed out that the effectiveness of the proposed approach is only verified via simulations.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
This work was financially supported by the Shanghai University of Engineering Science Development Fund to cultivate high-level special project (2012gp02). The authors would like to thank the editor and the reviewers for their constructive comments and suggestions which improved the quality of the paper.
BarzaminiR.ShafieeM.DadlaniA.Adaptive generalized minimum variance congestion controller for dynamic TCP/AQM networks201235217017810.1016/j.comcom.2011.08.0102-s2.0-82955167858JingY.WangH.PanW.LiuX.Robust stability analysis for uncertain state and input delay TCP/AQM network systemsProceedings of the American Control Conference (ACC '08)June 2008Seattle, Wash, USA2643264710.1109/ACC.2008.45868912-s2.0-52449131590KellyF. P.MaullooA. K.TanD.Rate control for communication networks: shadow prices, proportional fairness and stability199849323725210.1057/palgrave.jors.26005232-s2.0-0032027247LowS. H.A duality model of TCP and queue management algorithms200311452553610.1109/TNET.2003.8152972-s2.0-0042641574HollotC. V.LiuY.MisraV.TowslyD.Unresponsive flows and AQM performance1Proceedings of the 22nd Annual Joint Conference of the IEEE Computer and Communications (INFOCOM '03)March-April 2003859510.1109/INFCOM.2003.1208661BigdeliN.HaeriM.Predictive functional control for active queue management in congested TCP/IP networks200948110712110.1016/j.isatra.2008.10.0012-s2.0-57649126335ManfrediS.di BernardoM.GarofaloF.Design, validation and experimental testing of a robust AQM control200917339440710.1016/j.conengprac.2008.09.0132-s2.0-58149490959ZhengF.NelsonJ.An H∞ approach to the controller design of AQM routers supporting TCP flows200945375776310.1016/j.automatica.2008.10.014MR25272612-s2.0-60449089099TsumuraK.HaraS.NakajimaA.H2 performance limitation of congestion controller for TCP/AQM network systemsProceedings of the 44th IEEE Conference on Decision and Control, and the European Control Conference (CDC-ECC '05)December 2005Seville, Spain6768677310.1109/CDC.2005.15832502-s2.0-33847224110SunJ.KoK.-T.ChenG.ChanS.ZukermanM.PD-RED: to improve the performance of RED20037840640810.1109/LCOMM.2003.8156532-s2.0-0042889219FiruziM. F.HaeriM.Adaptive generalized predictive control of active queue management in TCP networksProceedings of the The International Conference on Computer as a Tool (EUROCON 2005)November 2005Belgrade, Serbia6766792-s2.0-3384777624210.1109/EURCON.2005.1630020TanL.ZhangW.PengG.ChenG.Stability of TCP/RED systems in AQM routers20065181393139810.1109/TAC.2006.876802MR22487372-s2.0-33747820540BigdeliN.HaeriM.CDM-based design and performance evaluation of a robust AQM method for dynamic TCP/AQM networks200932121322910.1016/j.comcom.2008.10.0052-s2.0-57349138271MisraV.GongW. B.TowsleyD.Fluid-based analysis of a network of AQM routers supporting TCP flows with an application to REDProceedings of the ACM SIGCOMMConferenceSeptember 20001511602-s2.0-0033651669HollotC.MisraV.TowsleyD.GongW.On designing improved controllers for AQM routers supporting TCP flows3Proceedings of 20th Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM '01)2001Anchorage, Alaska, USAIEEE1726173410.1109/INFCOM.2001.916670FloydS.JacobsonV.Random early detection gateways for congestion avoidance19931439741310.1109/90.2518922-s2.0-0027647490AnickD.MitraD.SondhiM. M.Stochastic theory of a data-handling system with multiple sources19826181871189410.1002/j.1538-7305.1982.tb03089.xMR6853122-s2.0-0020191755MascoloS.Congestion control in high-speed communication networks using the Smith principle199935121921193510.1016/S0005-1098(99)00128-4MR18283992-s2.0-0033284136ZhengY. G.WangZ. H.Stability and Hopf bifurcation of a class of TCP/AQM networks20101131552155910.1016/j.nonrwa.2009.03.008MR26465682-s2.0-77950916189ChiassonJ.LoiseauJ.2007Berlin, GermanySpringerMascoloS.Smith's principle for congestion control in high-speed data networks200045235836410.1109/9.839966MR17567712-s2.0-0033733826WuS.OkuboS.WangD.Design of a model following control system for nonlinear descriptor system in discrete time2008444546556MR2459072ZBL1173.933872-s2.0-55349092672WangD.WuS.OkuboS.The state predictive model following control system for linear time delays20096218619110.1007/s11633-009-0186-42-s2.0-65449188325HollotC. V.MisraV.TowsleyD.GongW.Analysis and design of controllers for AQM routers supporting TCP flows200247694595910.1109/TAC.2002.1008360MR19148282-s2.0-0036600891HollotC. V.MisraV.TowselyD.GongW. B.A control theoretic analysis of REDProceedings of the INFOCOM 12th Annual Joint Conference of the IEEE Computer and Communications Societies2001Anchorage, Alaska, USA15101519MaramiB.HaeriM.Implementation of MPC as an AQM controller201033222723910.1016/j.comcom.2009.09.0012-s2.0-72249108592AzumaT.FujitaT.FujitaM.Congestion control for TCP/AQM networks using state predictive control200512591491149610.1541/ieejeiss.125.1491AzumaT.FujitaT.FujitaM.A design of state predictive H∞ congestion controllers for TCP/AQM networks20051810373375AzumaT.NaitoH.FujitaM.Experimental verification of stabilizing congestion controllers using the network testbedProceedings of the American Control Conference2005Portland, Ore, USA18411846JingY.ChenZ.DimirovskiG. M.Robust fuzzy observer-based control for TCP/AQM network systems with state delayProceedings of the American Control Conference (ACC '10)July 2010Baltimore, Md, USAMarriott Waterfront135013552-s2.0-77957800561