Aggregate traffic performance with active queue management and Drop from Tail

Active queue management (AQM) refers to a family of packet dropping mechani sms for router queues that has been proposed to support end-to-end congesti on control mechanisms in the Internet. In this paper, we examine the perfor mance of AQM mechanisms by varying two parameters: the queue size and the d ropping function. AQM flavors considered include "RED", the more recently p roposed "Gentle RED" and an additional mechanism we call "Gentle RED with i nstantaneous queue size". We use experimentation to analyze the performance of the AQM mechanisms ide ntified above on the aggregate traffic going through a congested router. Th e metrics used are: TCP goodput, TCP and UDP loss rate, queueing delay and consecutive loss probability. The AQM mechanisms are compared to Drop from Tail, the buffer management mechanism currently found in most operational r outers. The major observation is that AQM mechanisms have a minor impact on the agg regate performance metrics we observe. On the other hand, we observe an imp ortant sensitivity of the AQMs considered to traffic characteristics that m ay compromise their operational deployment.