Emulated weighted fair queueing algorithm for high-speed integrated service networks

Abstract

Broadband integrated services packet networks rely on traffic scheduling algorithms within individual switches or routers to provide a wide range of Quality-of-Service (QoS) guarantees. The function of a scheduling algorithm is to select, for each outgoing link of the switch, the packet to be transmitted in the next cycle from the available packets belonging to the flows sharing the output link. Weighted Fair Queueing (WFQ) is known as an ideal traffic scheduling algorithm in terms of its delay and fairness properties. However, timestamp computation in the WFQ scheduler serving N sessions has a complexity of O(N) per packet-transmission time which makes its implementation difficult. There have been efforts in the past to simplify the implementation of WFQ, such as Self-Clocked Fair Queueing (SCFQ), Starting Potential Fair Queueing(SPFQ), Minimum Delay Self-Clocked Fair Queueing (MD-SCFQ), etc. SCFQ is simpler and more feasible for high-speed implementation than WFQ since it is based on an internally generated virtual time, but it has unbounded delay property due to its poor isolation properties among sessions; SPFQ achieves the same delay bounds of WFQ and comparable fairness index to WFQ, but it needs selecting a new minimum virtual start time every time a packet completes service, thus requiring another sorting procedure according to the virtual start time of the packets and has a virtual time computation complexity of O($\log N$); MD-SCFQ also has the same delay bounds of WFQ and comparable fairness index to WFQ, but it needs to keep some information and perform additional calculations to obtain the weighted average virtual start time of packets in the head-of-line of each queue. In this thesis we propose a new and efficient fair queueing algorithm, called Emulated Weighted Fair Queueing (EWFQ), which has a virtual time computation complexity of O(1), while it emulates perfectly the delay and fairness properties of WFQ. Key idea of EWFQ is that we reca...