My dissertation research project consists on applying the signal processing techniques to the problem of congestion control in computer networks. I am currently developing and implementing an AQM scheme, which outperforms previous algorithms (RED, REM, PI and AVQ) in detection and reaction speed, while maintaining the queue to a desired value.
The main principles include applying the likelihood ratio test for congestion detection based on an estimation of congestion by using a single queue occupancy sample. The packet marking/dropping probability is mapped using a simple function of the congestion measurement. Later, an algorithm is implemented to optimally spread packets far apart and reduce the queue variability.
Publications about this topic have been submitted to Transactions on Networking. The first publication refers to the Detection Problem, and which is attacked by using a likelihood ratio test based detector. The important results on this publication include the approximation of the queueing process to a Gamma distribution, based on this approximation, the congestion is shown to increase super-exponentially with respect to the queue occupancy. However, the general form to obtain the likelihood of congestion is open to different distributions, and it is shown that a Gamma distributed queue fits better the simulated results when compared to Gaussian and Uniform distributions
The second publication corresponds to a novel AQM method based on the likelihood detector and the half-toning technique called Error Diffusion. Error difussion allows to spread packets far appart with good precision. The two thecniques mixed in a weighted system to place marks, shows to considerably reduce the queue variability while keeping the bottleneck link utilization high for wide spectrum of network scenarios.
The publications in PDF form can be found at: Statistical Detection of Congestion in Routers and Statistical Approach for Congestion Control in Gateway Routers.
If you want to know more about my views on congestion control click HERE