Discussion of explicit congestion notification's impact on the Internet
 help / color / mirror / Atom feed
* [Ecn-sane] cwnd/2 - 1
@ 2018-09-07  7:18 Dave Taht
  2018-09-10 11:26 ` Pete Heist
  0 siblings, 1 reply; 2+ messages in thread
From: Dave Taht @ 2018-09-07  7:18 UTC (permalink / raw)
  To: ecn-sane

One of the things that went by in recent kernel changes was that
linux tcp at a cwnd 2 on a CE goes into quickack mode, so that we
get back one ack per packet (I think, from reading the commit) instead of
every other one.

halving cwnd and then taking off 1 more gives a response that drops the
mass closer along the lines of what a drop may do in response to loss
and re-ordering. A cwnd 1 in quickack mode would halve the number of
packets outstanding in my 100 flows test. I think.

another thing to note is that retransmits, are, by the standard, and
in linux, not protected by ECT, and when you get near to 100% mark
rate as i do in this artificial test, get dropped a lot.



-- 

Dave Täht
CEO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-669-226-2619

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2018-09-10 11:26 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-09-07  7:18 [Ecn-sane] cwnd/2 - 1 Dave Taht
2018-09-10 11:26 ` Pete Heist

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox