Weighted round robin (WRR) is a simple, efficient packet scheduler providing low latency and fairness by assigning flow weights that define the number of possible packets to be sent consecutively. A variant of WRR that mitigates its tendency to increase burstiness, called interleaved weighted round robin (IWRR), has seen analytical treatment recently \cite{TLBB21}; a network calculus approach was used to obtain the best-possible strict service curve. From a different perspective, WRR can also be interpreted as an emulation of an idealized fair scheduler known as generalized processor sharing (GPS). Inspired by profound literature results on the performance analysis of GPS, we show that both, WRR and IWRR, belong to a larger class of fair schedulers called bandwidth-sharing policies. We use this insight to derive new strict service curves for both schedulers that, under the additional assumption of constrained cross-traffic flows, can significantly improve the state-of-the-art results and lead to smaller delay bounds.
翻译:环绕龙虾(WRR)是一个简单、高效的包装调度器,通过分配流量重量来界定可能连续发送的包包数量来提供低延缓度和公平性。WRR的变式减轻了它增加防暴性的趋势,称为间断加权环绕龙虾(IWRR),最近进行了分析处理;网络微积分法用于获得最可能达到的严格服务曲线。从不同角度看,WRR也可以被解释为模拟一个理想化的公平调度器,称为普遍处理共享(GPS ) 。在GPS业绩分析的深刻文献结果的启发下,我们显示WRR和IWRR都属于一个叫作带宽共享政策的更大型的公平调度器。我们利用这一洞察力为两个调度器提供新的严格的服务曲线,根据受限制的跨贸易流动的额外假设,这可以大大改进状态结果,并导致更小的延迟界限。