Deterministic fluid models for internet congestion control
暂无分享,去创建一个
In this thesis, we use deterministic fluid models as mathematical tools to study various aspects of designing congestion control mechanisms for the Internet.
We first consider the problem of congestion control in networks with both multirate multicast sessions and unicast sessions. We present a decentralized algorithm that enables the different rate-adaptive receivers in different multicast sessions to adjust their rates to satisfy some fairness criterion. A one-bit ECN-marking strategy to be used at the nodes is also proposed. We also prove the stability of the proposed mechanism.
We then consider a single bottleneck link accessed by a single congestion-controlled source. We derive conditions for the semiglobal exponential stability of such congestion controllers. Using numerical examples we show that, the condition for local non-oscillatory stability is sufficient for global asymptotic stability for reasonable marking functions.
Next we validate the use of deterministic delay differential equation models for congestion control. We consider the case of a single link accessed by many TCP-like congestion-controlled flows, and uncontrolled flows which are modeled as stochastic disturbances. We show that if the number of flows is large and the link capacity is scaled in proportion to the number of users, then, the trajectory of the stochastic system is eventually well approximated by the trajectory of a delay-differential equation. We show that a system with multiple TCP-like flows is globally stable (and thus, a deterministic model is reasonable) as long as product of the throughput and feedback delay per flow is not very small.
We then study the impact of parameter choice of an AQM (active queue management) scheme that can be used at the router. We argue that, depending upon the choice of the parameters of the AQM scheme, one would obtain a rate-based model or a rate-and-queue-based model as the deterministic limit of a stochastic system with a large number of users. We argue that, a virtual-queue-based AQM scheme is very robust to the choice of parameters in achieving a low-loss, low-delay, and high-utilization operation. However, with a real-queue-based marking, the choice of parameters is much more critical.