FGFFQ: A Scalable Real-Time Packet Scheduling Technique for Terabit Routers

Weighted Fair Queuing (WFQ) is a packet scheduling technique that has been proposed and is currently used in modern network routers to provide end-to-end delay and bandwidth guarantees. Unfortunately the scalability of existing WFQ implementation techniques in large-scale backbone routers is relatively poor. Various approximation techniques have been proposed to improve WFQ's scheduling eeciency at the expense of looser performance bounds. This paper describes a packet scheduling technique called Fixed-Granularity Fluid Fair Queuing (FGFFQ), which is shown analytically and empirically to be more scalable than WFQ. The key idea in FGFFQ is to transparently add a simple and eecient multiplexing/demultiplexing mechanism above existing link-layer protocols so that network-layer packets do not have to be travel on network links in contiguous pieces. FGFFQ not only can completely eliminate WFQ's packet priority calculation and sorting overhead, but also can approximate the uid fair queuing model by choosing appropriate scheduling granularities. In addition, the scheduling model of FGFFQ supports both work-conserving and non-work-conserving disciplines and can accommodate low-latency and low-bandwidth real-time connections such as Internet telephony applications without wasting network bandwidth resources.