Abstract: Compound TCP (CTCP) is a practical congestion control algorithm for high-speed and long-delay networks. This paper addresses the problem of bifurcation and chaos analysis and its control in a nonlinear congestion control model with CTCP connections and a Random Early Detection (RED) gateway. A hybrid control strategy using both state feedback and parameter perturbation is employed to stabilize chaotic orbits embedded in this discrete-time dynamic system. Theoretical analysis and numerical simulations show that the controlled system achieves a stable average queue size over an extended range of parameters, effectively decreasing the sensitivity of RED parameters and eliminating chaotic behavior.
The Problem: Nonlinear Dynamics and Network Chaos
In high-speed, long-delay networks, the Transmission Control Protocol (TCP) manages how data is sent, while routers use Active Queue Management (AQM) like Random Early Detection (RED) to prevent congestion. RED works by monitoring the average queue size and dropping packets early to signal the sender to slow down.
However, the interaction between the protocol (CTCP) and the router (RED) is nonlinear. If system parameters—like the weight ($w$) used to calculate average queue size—are tuned incorrectly, the system can undergo bifurcation. This means the queue size stops being steady and starts oscillating between multiple values, eventually collapsing into chaos.
As shown in the bifurcation diagram above, the system enters chaos through a "period-doubling" route. Chaotic behavior leads to unpredictable delays and degraded network performance, making RED notoriously difficult for network administrators to tune.
Methodology: A Hybrid Control Strategy
To address this instability, the authors developed a hybrid control strategy that combines two distinct techniques:
- State Feedback: Using the current state of the queue to influence the next step in the calculation.
- Parameter Perturbation: Making small, calculated adjustments to system parameters to nudge the system back toward stability.
The researchers utilized a discrete-time dynamic model to represent the CTCP/RED system. A key advantage of this hybrid controller is that it is "non-invasive"; it stabilizes the system to the same desired average queue size (the fixed point) as the original model, but without the chaotic fluctuations. It essentially increases the "basin of attraction," making the stable state much harder to disrupt.
Key Results and Stabilization
The effectiveness of the hybrid controller was validated through numerical simulations using standardized network parameters (e.g., 15 Mbit/s link capacity, 4000-bit packet size).
One of the most significant findings was the controller's ability to delay the onset of bifurcation. In the original uncontrolled system, the critical threshold for the averaging weight ($w$) was $0.269$. With the hybrid controller ($r=0.8$), this threshold was pushed to $0.336$, significantly widening the "safe" operating range for the network.
As illustrated in the time-series simulation, when the controller is activated during a chaotic state (at $t=200$), the average queue size stabilizes to the fixed point almost instantly. The research also demonstrated that the controller is robust across various parameters, including the number of TCP connections ($N$) and propagation delay ($d$). For certain parameters like maximum drop probability ($p_{max}$), the strategy was able to eliminate chaotic behavior entirely across specific ranges.
Conclusion
This study provides a mathematical framework for stabilizing high-speed networks against the inherent instabilities of nonlinear congestion control. By applying a hybrid control strategy, network administrators can decrease the sensitivity of RED parameters, ensuring predictable performance even in complex, high-speed environments. Future work may explore delayed feedback control strategies to better account for the inherent lags in real-world network hardware.
Summary updated