Push forward link-level scheduling for network-wide performance

Leandros Tassiulas

Abstract: A virtual circuit network with arbitrary topology is considered. The traffic streams follow prespecified routes, different in general for each stream, to reach their destination. A fluid traffic model is adopted and a processor sharing service discipline is considered. A policy is proposed for setting adaptively the fractions of the transmission capacity, which is allocated to the different traffic streams in the processor sharing discipline at each link. The amount of traffic arrived at the originating node of each link is measured for each stream. The fraction of the link capacity allocated to each stream is set to be proportional to the measured traffic. The traffic is measured continuously and the fractions are updated regularly based on the most recent traffic measurements. It is shown that eventually, the transmission capacity allocated to each stream converges to a quantity proportional to the average rate of the stream. Hence, if the capacity condition is satisfied, sufficient fractions of the capacity are allocated at each link for each stream. End-to-end performance guarantees are provided, if the traffic is regulated. The policy is distributed since each link adjusts the service fractions based on observations of the traffic arriving at its originating node only. Furthermore, it is adaptive since no information on the traffic characteristics is needed for the application of the policy.