A unified approach to ameliorate active queue management of network routers

Abstract

The layering technique has divided TCP/IP protocols into a few layers and assigned each layer with some special functionality. The mechanism of flow and congestion control in TCP at the transport layer can be dispersed to IP at the network layer where there used to be a simple strategy (ie. drop-tail) of managing the router's buffer. Active queue management can improve network efficiency by detecting impending congestions in Internet routers. The first such effort was random early detection (RED) introduced in the early 1990s, and a number of active queue management algorithms have been developed since then. The essence of AQMs is to adjust the packet dropping rate before the buffer becomes completely full. The dropping probability is a determining factor in affecting the overall throughput and transmission delay. Various AQM algorithms have been developed to produce a just-right dropping rate largely based on the statistical data. Unfortunately, all these algorithms possess some drawbacks in one way or another. In this paper, we devise a unified approach to determining the packet dropping possibility, which is sensitive to all factors that might have some ultimate effect on the buffer's occupancy. The proposed approach has been tested on a simple network using OMNET++ [A. Varga].