OWFQ: Reducing Packet Drops for Approximate Weighted Fair Queueing with Calendar Queues

Weighted fair queueing (WFQ) is a router mechanism that can achieve customized bandwidth allocation and flow isolation in data centers as well as other high-speed networks today. However, it’s hard to faithfully implement the original WFQ algorithm in high-speed networks due to the high complexity such as the need for sorting packets at line rate. Therefore, many approximate WFQ packet schedulers have been proposed as a substitute, such as PCQ, a packet scheduler that uses calendar queues. Unfortunately, although PCQ successfully achieved fairness among flows without such complex hardware requirements, it introduced a new problem that the packet drop rate is significantly increased when compared with ideal WFQ algorithm as a result of its implementation mechanism and approximation cost. As we all know, packet drop is a common but important problem. High drop rate not only affects bandwidth utilization but also has a negative effect on the QoS (quality of service) of latency-sensitive services. In this paper, we design a new approximate WFQ scheduling algorithm with calendar queues named opportunistic weighted fair queueing (OWFQ). OWFQ is designed to reduce packet drops caused by the approximation cost of calendar queues. We conduct simulations on an incast network with one switch to measure drop rates of different algorithms. The results show that OWFQ significantly decreases the drop rate when compared with PCQ and thus improves bandwidth utilization.