130 likes | 279 Views
A CTIVE Q UEUE M ANAGEMENT. Comparing RED and BLUE Presentation held at: Telia Prosoft AB, Farsta, Sweden 2001- May-18 Thesis worker: Bart ók István bartoki@hpn.telialabs.net Consultant: Imre Juhász Imre.I.Juhasz@telia.se. Introduction. Demand for short Round Trip Time
E N D
ACTIVE QUEUE MANAGEMENT Comparing RED and BLUE Presentation held at: Telia Prosoft AB, Farsta, Sweden 2001-May-18 Thesis worker: Bartók István bartoki@hpn.telialabs.net Consultant: Imre Juhász Imre.I.Juhasz@telia.se
Introduction • Demand for short Round Trip Time • Wish for low packet loss rate • All this with best effort traffic • “Guaranteed minimum + fair share” broadband package customers will saturate any backbone
The Challenge • Drop-tail is not adequate • Random Early Detection is not perfect • What about Explicit Congestion Notification? • Many proposed algorithms to replace RED • Is BLUE the right one? • Add BLUE to the Linux kernel • Compare by measurements
Drop-tail • Signals congestion only when the queue is already saturated • Likely to drop more packets from the same flow • Unfair with bursty flows
The problems with RED • The queue length either oscillates, • Or it cannot control the congestion
Other Algorithms • Adaptive RED • P-I Controller • Adaptive Virtual Queue • BLUE • ...
BLUE • Tail drop happens? Increase Pdrop • Queue empy? Decrease Pdrop • Freeze Pdrop after change • Its author’s simulations are promising:
Linux kernel Traffic Control • Nice Framework • Part of the standard Linux kernel • But written by a Real ProgrammerTM Bare documentation
Real Life • ECN problems • Retransmits • SYN • High bias towards ECN when naive • Standardization problems - Linux & FreeBSD ECN breaks the RFC • Take the BLUE pill?
Conclusion • The perfect mechanism still has not been arrived • Design for Nonlinear • Include Arrival & Departure Rate
Future Work • Measure & Simulate • Improve some of the algorithms, or • Design The One • Good background is provided in the thesis doc: • Guide to TCP Congestion Control • Guide to Linux kernel Traffic Control Framework • Evaluation of some Queue Management algorithms