dAQM: Derivative-Based Active Queue Management

The network systems build heavily on Active Queue Management(AQM) algorithms for maintaining an optimal queue size and avoiding issues like Bufferbloat. Despite promising performance, the current AQM algorithms face a major challenge of estimating the accurate congestion due to bursty network traffic. The major reason is the use of baseline traffic features like average delay and sojourn time. In this paper, we propose a novel AQM algorithm, called dAQM, which uses advanced traffic features like three higher-order derivatives of sojourn time and buffer size for computing the packet drop probability based on the network congestion. The higher-order derivatives of sojourn time and buffer size provide the rate of increase of packet accumulation in the queues. We show the programmability and performance analysis of dAQM over five traffic distribution models(Pareto, Poisson, etc.) for five widely used traffic classes. The analysis over ns-3 simulations showed that dAQM provides at least 28% and 62% improvement in delay and queue length, respectively, for an increase in traffic load as compared to the traditional AQM algorithms. dAQM also reduces the flow completion time for long flows, i.e., FTP traffic, by at least 39.7% as compared to the prior AQM algorithms.