API: adaptive proportional-integral algorithm for active queue management under dynamic environments

This paper studies active queue management (AQM), and proposes an adaptive proportional-integral (API) algorithm for AQM, based on an existing PI algorithm. API alms to expedite the control response through online measurement and adjustment. First, API needs to periodically measure average queue length and packet loss ratio; second, it can dynamically adjust the packet drop probability using the measured results. Due to the use of online measurement and adjustment, the queue length under API can be much more quickly converged to around the desired value q₀, and with a faster control response than PI, especially under heavy congestion. API is well suited to dynamic network environments where traffics change frequently and it is hard to predict them. The simulation results validate this advantage. Although API introduces online measurement and adjustment, it still maintains O(1) complexity and can be efficiently deployed in high-speed routers.