Re: Mice [was Re: QoS for system critical packets on wireless]

[Dave Taht <dave.taht@gmail.com>]

OK, so I've gone and updated the related wiki page to indicate more
of what I'd actually meant. Still need a good analogy though...

Secondly you've given me an idea towards actively tracking mice-y
short flows using something like conntrack or simply getting stats on
syns, synacks,fins, finacks which ought to be interesting in itself.
One problem I'd already found in the openwrt defaults was that it does
severe rate limiting of syns by default - 25!!!??? which I only found
by trying to access the top 100 web sites all at the same time and
being totally befuddled by the results. (this test is also in my
Diffserv repo)

Thirdly, although the iptables implementation of diffserv was
straightforward and
easy to describe, and the proposal for enhancing iptables to be
performant equally clear and straightforward, it's beginning to appear
that most of the functionality I proposed in the first draft of the
diffserv rfc belongs in the 'tc' subsystem, not in iptables.

I'm really glad I wrote it down and am grateful for the feedback both
public and private thus far.


-- 
Dave Täht
SKYPE: davetaht
US Tel: 1-239-829-5608
http://the-edge.blogspot.com