The recent studies show that the network traffic is self-similar.This nature has a great impact on flow control and queuing performance in networks

therefore it has received significant attention.By taking advantage of self-similarity of network traffic

we propose a new algorithm

named DMPIC-AQM (Dynamic Matrix PI Control -Active Queue Management).This algorithm has two significant features.On one hand

it makes use of linearity regression theory to predict the characters of future traffic.On the other hand

it is capable to control the active queue based on the principle of predictability PI controller in control theory.Thus

the algorithm can both restrict the oscillations and stabilize the length of queue in expectation.The simulation experimental results show that the integrated performance of the proposed algorithm is obviously superior to RED and PI algorithms

especially when network traffics change dramatically and the length of queue is short.