Discretization in Fluid-Fairness : Formulation and Implications

This work analyses the implications of discretization on fluid-fairness. In this context, discretization means being fluid-fair with a chosen granularity. The operation of a discretized scheduler is very simple and has constant overhead, and provides a parameterizable deviation from the optimal delay and fairness bounds provided by a fluid scheduler. Discretized schedulers can actually improve the end-to-end delay bounds by achieving a cutthrough forwarding effect through a network path. We derive these deviations for both work-conserving and nonwork-conserving discretized schedulers. The traffic model used in the analysis is the one proposed in [2]. We also present an operational overhead analysis comparing packetized and discretized schedulers. Studying discretization is interesting since most high-speed switch cores use at least one notion of discretization, viz, discretization in size. Moreover, discretization permits efficient hardware implementations for scheduling and processing flows.