Buffer management schemes for supporting TCP in gigabit routers with per-flow queueing

Recently, there has been much interest in using active queue management in routers in order to protect users from connections that are not very respon sive to congestion notification. A recent Internet draft recommends schemes based on random early detection for achieving these goals, to the extent t hat it is possible, in a system without "per-flow" state. However, a "state less" system with first-in/first-out (FIFO) queueing is very much handicapp ed in the degree to which flow isolation and fairness can be achieved, Star ting with the observation that a "stateless" system is but one extreme in a spectrum of design choices and that per-flow queueing for a large number o f flows is possible, we present active queue management mechanisms that are tailored to provide a high degree of isolation and fairness for TCP connec tions in a gigabit IP router using per-flow queueing. We show that IP flow state in a router can be bounded if the scheduling discipline used has fini te memory, and we investigate the performance implications of different buf fer management strategies in such a system. We show that merely using per-f low scheduling is not sufficient to achieve effective isolation and fairnes s, and it must be combined with appropriate buffer management strategies. R ecently, there has been much interest in using active queue management in r outers in order to protect users from connections that are not very respons ive to congestion notification. A recent Internet draft recommends schemes based on random early detection for achieving these goals, to the extent th at it is possible, in a system without "per-flow" state. However, a "statel ess" system with first-in/first-out (FIFO) queueing is very much handicappe d in the degree to which flow isolation and fairness can be achieved, Start ing with the observation that a "stateless" system is but one extreme in a spectrum of design choices and that per-flow queueing for a large number of flows is possible, we present active queue management mechanisms that are tailored to provide a high degree of isolation and fairness for TCP connect ions in a gigabit IP router using per-flow queueing. Me show that IP flow s tate in a router can be bounded if the scheduling discipline used has finit e memory, and we investigate the performance implications of different buff er management strategies in such a system. We show that merely using per-fl ow scheduling is not sufficient to achieve effective isolation and fairness , and it must be combined with appropriate buffer management strategies.