Maxwell strikes the heart (ECN: Clearing the air)
Quoted 11 lines
From: Gregory Maxwell ([email protected])
Date: Sun Jan 28 2001 - 14:42:04 ESTQuoted 6 lines · James Sutherland
Quoted 2 lines
There is nothing silly with the decision, davem is simply a modern day internet hero.
No. If it were something essential, perhaps, but it's just a minor performance tweak to cut packet loss over congested links. It's not IPv6. It's not PMTU. It's not even very useful right now!
Quoted 7 lines
No. ECN is essential to the continued stability of the Internet. Without probabilistic queuing (i.e. RED) and ECN the Internet will continue to have retransmit synchronization and once congested stay congested until people get frustrated and give it up for a little bit.
It's a real issue, and it's actually important to have it implemented. It's not just a performance hack.
I always "knew" that the stability of the Internet is secured by the exponential backoff in TCP. A small packet loss on uncongested links is a part of this technique, and it existed long before ATM studies produced RED (which infiltrated backwards). It also requires sending stacks to "give up for a little bit" (actually to give up a lot, and together with the slow start it produced the well known "saw" of the window size).
So far I fail to see how a repainted NAK, kludged into a NAKless protocol, would improve stability of the Internet. If anything, it is going to exaggerate traffic oscillations. I would appreciate couple of links to reputable studies or discussions on the subject.
-- Pete
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/
Quoted 3 lines · Pete Zaitcev
So far I fail to see how a repainted NAK, kludged into a NAKless protocol, would improve stability of the Internet. If anything, it is going to exaggerate traffic oscillations.
If anything RED will *reduce* oscillations by alleviating retransmit synchronization.
Quoted 2 lines
I would appreciate couple of links
to reputable studies or discussions on the subject.
See
http://www.aciri.org/floyd/ecn.html
http://www.aciri.org/floyd/red.html
(I particularly like ftp://ftp.ee.lbl.gov/talks/vj-nanog-red.pdf)
RED and ECN can be considered independently (although they do work together quite well). You can see ECN as a way of the network telling the endpoints "Normally I would have dropped this datagram, but I'll let you get away with it this time". TCP behavior on reception of an ECN should be the same as when a datagram is dropped, but without the latency (and jitter) that is the result of a lost datagram.
JDB.
[using RED/ECN-like protocols for latency sensitive video communications over a wireless link]
Signature · 6 lines
LART. 250 MIPS under one Watt. Free hardware design files. http://www.lart.tudelft.nl/ - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] Please read the FAQ at http://www.tux.org/lkml/