Speeding up the Convergence of Estimated Fair Share in CSFQ

P. Wang and D.L. Mills (USA)


CSFQ, Fairness, Convergence, Network Management


Core-stateless Fair Queueing (CSFQ) is a scheme to achieve approximate fair bandwidth sharing without per flow state in the interior routers. The extra packets that beyond the fair share for each flow are dropped proba bilistically based on the attached flow rate in the packet header and the estimated fair share. A heuristic method is used to estimate the fair share in CSFQ. In our previous work CSFQIMP (CSFQ Improvement), we took the prob abilistic idea from SRED (Stabilized RED) and applied it in CSFQ to estimate the fair share. The probabilistic ap proach achieves a comparable or even better performance than the original heuristic approach. However, the conver gence speed of the probabilistic approach is slow. There fore, we propose a method to speed up the convergence of the fair share estimate in this paper. The idea comes from that the router randomly selects k packets instead of one packet to compare with the incoming packet in SRED. We show that the convergence speed is increased in the usual cases. Simulation results show that the speedup approach achieves a quick convergence compared with the original probabilistic method in our previous work.

Important Links:

Go Back