Link-level scheduling for network-level performance

A service discipline for the link transmission scheduling in a virtual circuit network with arbitrary topology is proposed. The scheduling of each link is based on the traffic at its origin node only. It is performed in two levels operating in different time scales. In the slower time scale the fraction of the link capacity to be allocated to the different streams is updated such that it remains proportional to the traffic of each stream that has arrived at the origin node of the link. In the faster time scale the link capacity is shared, such that the fractions of the capacity allocated by the bandwidth allocation policy are honored. A fluid traffic model is considered and processor sharing service is assumed. The policy is analyzed and it is shown that the existence of rates of the arrival streams guarantees that the allocated bandwidth at each stream will converge to a value larger than the link load. When the arrival streams are regulated to satisfy certain burstiness constraints, then the backlog at each network node is bounded as well