Block-based fair queuing: An efficient network QoS provisioning algorithm for high-speed data transmission

We propose an extension of WFQ, called BWFQ, to reduce the computation time and guarantee the QoS performances as well. Simplifications of the WFQ algorithm may degrade the performance of all sessions. Instead, BWFQ dynamically aggregates packets in advance and performs scheduling in unit of block. Block size is controlled by a parameter called granule. Each session has a granule setting in accordance with its delay tolerance. Delay-insensitive sessions form larger data blocks so that a lot of computation time at scheduling is saved. BWFQ also employs block-preemptive scheduling to allow interruption of the block on service and switch to another one with the highest priority. Performance isolation guarantees a delay upper bound for every session that is independent of others' granule settings. Simulation results show that BWFQ is capable of saving a great amount of computation, especially under traffic congestion.