On the deterministic approach to active queue management

Abstract

Virtually all known active queue management (AQM) algorithms, except for the two-category classifier (TCC), operate by calculating packet dropping probabilities. The probabilistic approach involves the necessity of using a type of random number generation upon every packet arrival at the router. Even if the generation of a single random value does not involve high computational complexity, the overhead becomes significant, considering the number of packets in a typical Internet environment. We propose a new AQM algorithm based on the deterministic approach. The algorithm offers a high throughput and a low loss ratio while maintaining a short and stable queue size. At the same time, the algorithm is of low computation complexity, which allows for energy-efficient implementations in routers. In addition to that, the proposed algorithm is universal--it provides high performance in a variety of distinct networking scenarios (diversified round-trip times, congestion levels, traffic types, etc.). Contrary to the TCC algorithm, the proposed algorithm does not make the decision whether to accept or drop the packet upon each packet arrival--the computation process is involved less frequently.