A practical fair queuing scheduler: Simplification through quantization

Abstract

The design of fair packet schedulers involves a tradeoff between implementation complexity, on one hand, and delay and fairness guarantees, on the other. In this paper, we present tiered-service fair queuing (TSFQ), a new scheduler that exploits certain properties of Internet traffic to speed up the bottleneck operations related to virtual time computation and packet sorting. Specifically, TSFQ makes innovative use of quantization (along the two dimensions of flow weights and packet lengths) her with specialized yet simple queuing structures. TSFQ combines all three properties that are important to a fair queuing algorithm, namely, a tight delay bound, worst-case fairness, and low complexity and amenability to hardware implementation. Hence, we believe that, for network operators, deploying TSFQ scheduling has the potential to enhance their ability to offer and guarantee a wide range of services.