Why WFQ Is Not Good Enough For Integrated Services Networks

Weighted Fair Queueing has received a lot of attention in the research community recently. In fact, the IETF has proposed to use WFQ as a basic building block for future integrated services networks. The reason for such a great interest is that it closely approximates a hypothetical uid model discipline called Generalized Processor Sharing (GPS), which provides a conceptually ideal building block for integrated services network. There is a general misconception that WFQ provides almost identical service to GPS with a maximum diierence of one packet, thus it is an ideal and practical replacement for GPS. This is a misinterpretation and over-generalization of Parekh's result that the delay bound provided by WFQ is within one packet transmission time of that provided by GPS. In this paper, we will show that, contrary to popular belief, there could be large discrepancies between the services provided by the packet WFQ system and the uid GPS system. The inaccuracy introduced by WFQ in approximating GPS are detrimental to both real-time and best-eeort traac in a network with hierarchical link-sharing service. In addition, WFQ has a relatively high complexity which make it diicult to implement. We propose two new algorithms and show that they overcome the limitations of WFQ. Simulation results are presented to demonstrate the advantages of the new algorithms.