AFRED: an adaptive fuzzy-based control algorithm for active queue management

This paper studies the active queue management (AQM) in high-speed routers. One of the original and popularly implemented AQM algorithms is random early detection (RED). But it is hard to configure the parameters involved in RED. Although there are some new TCP-modeling based approaches to overcome this, they really show some instability under such dynamical environments with diverse connections (adaptive or non- adaptive, short or long), link rate, RTT, and buffer size. The major reason is that these approaches are primarily based on TCP model, which has some mismatches under dynamical environment. Since fuzzy-logic control does not need precise control model, it may be suitable for such environment. This paper designs an adaptive fuzzy-based control algorithm (AFRED), which computes the packet drop (or marking) probability according to pre-configured fuzzy logic using the instant queue length as input variable. The important feature of AFRED is that it also designs an adaptive mechanism to dynamically readjust the fuzzy rule so as to make AFRED itself extensively stable for many dynamic environments. Simulation results show that AFRED can effectively and stably control the queue length to the expected value or so. Compared with RED and PI algorithm, AFRED obtains higher goodput and stable queue length than RED and PI, even if introducing UDP flows.