Fair Bandwidth Sharing under Random Flow Arrivals/Departures: Metastability & Flow Admission Control
Vladimir V. Marbukh
Flow-level performance of fair bandwidth sharing protocols under random flow arrivals/departures is not sufficiently understood due to intractability of the corresponding Markov model for realistic-size networks. Tractability of the Markov model in fluid asymptotic regime presents an opportunity for addressing basic performance issues, including stability. This paper?s contributions are as follows. The paper proposes a flow-level Markov model of fair bandwidth sharing, which takes into account that the file transfer rates are determined by the corresponding good-puts rather than throughputs. The paper shows that since good-puts decrease as the numbers of flows in progress increase, this more realistic Markov model is typically unstable. Under fluid asymptotic regime, the paper demonstrates that despite this instability, a desirable metastable lightly loaded network mode of operation may still exist. The network may be stabilized in a close neighborhood of this desirable state, i.e., this metastable state can be transformed into stable state, with the appropriately designed flow admission strategy, while maintaining asymptotically small flow rejection probability.
International Symposium on Information Theory and Applications | 2007 |
June 24-27, 2007
IEEE International Symposium on Information Theory 2007
Fair Bandwidth Sharing under Random Flow Arrivals/Departures: Metastability & Flow Admission Control, International Symposium on Information Theory and Applications | 2007 |, Nice, FR, [online], https://doi.org/10.1145/1330555.1330560
(Accessed December 11, 2023)