Queue management algorithms and network traffic self-similarity

The self-similarity of network traffic has been established in a variety of environments and it is well known that self-similar traffic can lead to larger queueing delays, higher drop rates and extended periods of congestion. In this paper, we investigate the impact of various buffer management algorithms on the self-similarity of network traffic. In this paper we investigate the impact of active and passive queue management policies used at the routers on the self-similarity of TCP traffic. We also propose a modification to the random early detection (RED) algorithm, aimed at reducing the timeouts and exponential backoffs in TCP flows, and show that it can lead to significant reductions in the traffic self-similarity under a wide range of network conditions, as compared to the currently implemented active and passive buffer management policies. We also show that though our techniques are aimed at TCP related causes, it is also effective in reducing the degree of self-similarity in traffic even when application and user level causes are also present, as long as TCP is used as the underlying transport protocol.