Anti-windup Compensation in TCP/IP Routers: A Multi-delay Feedback Systems Approach
The design of anti-windup compensators for TCP/IP congestion in the presence of multiple classes of traffic is addressed here: a methodology is presented that takes into account the constraints to achieve the desired queue size, guaranteeing the stability
- PDF / 892,730 Bytes
- 14 Pages / 439.37 x 666.142 pts Page_size
- 96 Downloads / 146 Views
Abstract. The design of anti-windup compensators for TCP/IP congestion in the presence of multiple classes of traffic is addressed here: a methodology is presented that takes into account the constraints to achieve the desired queue size, guaranteeing the stability even in the presence of disturbances. Our proposed technique is based on using LMIs that include a tuning parameter, which makes possible to enlarge the domain of attraction to ensure the stability in the varying conditions inherent to TCP/IP traffic. This is so far the first control theory based approach for anti-windup in router that allows multi class Active Queue Management, considering delays, input saturation and variable link bandwidth (modeled as a time-variant disturbance). These characteristics are relevant for framing the network within Service Level Agreements, as data traffic is not homogeneous, and transmission delays are variable. Keywords: Anti-windup Congestion Domain of attraction Control theory
1 Introduction The increased demand for the Internet to transmit time-sensitive voice and video applications require new congestion control algorithms that take into account flows of different characteristics when regulating the demand of the network’s resources. Active Queue Management (AQM) - based algorithms are frequently used in routers to inform TCP’s senders about congestion, so senders fit their sending rate to the network characteristics. This is inherently a feedback control issue, so significant research is network characteristics. This is inherently a feedback control issue, so significant research is been devoted to the use of control theory to develop more efficient congestion control algorithms. A simple one is Drop Tail, which acts by dropping packets © Springer Nature Switzerland AG 2019 Y. Farhaoui and L. Moussaid (Eds.): ICBDSDE 2018, SBD 53, pp. 190–203, 2019. https://doi.org/10.1007/978-3-030-12048-1_21
Anti-windup Compensation in TCP/IP Routers
191
arriving at a router when the buffer is full. As emphasized in [1], this leads to flowsynchronization and performance degradation due to excessive time outs and restarts. As an improvement, RED was proposed [2–4]: it uses the EWMA of the queue length as a congestion indicator to drop/mark packets in order to achieve good balance between throughput and queuing delay. However, RED suffers from the lack of robustness with respect to tuning parameters which may easily lead to queue oscillation and instability. Different forms of controllers have been proposed in the literature based on Control Theory: for example, [5, 6] use the phase margin specification in the frequency domain to tune PI(D) controller gains. It must be pointed out that many of these Congestion Control methodology proposed in the literature based on Control Theory ignore the inherent time-varying delays, which is not adequate in practice: the packet drop/mark probability only has its effect on the senders after the Round Trip Time (RTT). It is then important to design controllers that do not ignore this delay, as
Data Loading...
