[Cerowrt-devel] linux 3.3 is out... but don't get excited

[Dave Taht]

http://lwn.net/Articles/487085/

I note that the linux release process is such that once a X.Y.0 comes
out, a lot of people start leaping on the release and a whole bunch of
patches show up. Usually usually X.Y.4 or so is truly stable, which I
figure will take about a month longer. This is about how long before
I plan to have a final 3.3 based release of cerowrt.

I'm doing a build of 3.3.0 from a clean toolchain as I write, but I
don't know if I will have time to test and release it before ietf.

In the final Linux 3.3 we did, at the last minute, have to revert one
of the key new sfq features, but the rest of the good stuff remains.

http://www.bufferbloat.net/issues/332

It's REALLY hard to hit it if sfq uses higher values (which is what
cerowrt uses, in combination with the randomness induced by red), but
it affected backward compatibility with older users of sfq in the
field, and perhaps a better answer exists, sooo... the next build of
cerowrt (3.3.0-1) will have this patch removed for a bit of testing
without and then I'll decide whether to add it back in, or fix it
somehow.

At the moment I see nothing in the next linux 3.4 release that is
relevant to bloat or for that matter ipv6. :(

I have some hopes that "tcp fast open" will show up in the 3.4
timeframe, and some hope van and kathies work may appear, but that's
it.
A bunch of new drivers with bql support, but not relevant to embedded.
I'd hope for more bql support in things like the dreamplug, too.

So I'm mostly thinking that after cerowrt 3.3... this will be a good
time to take a very long vacation, and I keep hitting reload here:

http://www.nicaraguasurfreport.com/reportlist.php?id_secc=25&x_date=2012-03-16&z_date=%3D%2C%27%2C%27

:)

maybe work directly on the mainline kernel for new stuff like iptables
enhancements, or try to make a dent in wireless, try to come up with
some better analytical software for analyzing streams, play with pcp,
hack the gui, I really don't know. We've just made a few breakthroughs
on the cosmic background bufferbloat detector project, too... so maybe
hack some hardware?

https://lists.bufferbloat.net/pipermail/thumbgps-devel/2012-March/thread.html

and I'm sore tempted to leverage concepts like that in the 'smile
plug' to build a 'perfect router'.

Suggestions wanted as to where to go after this release, feature
requests gladly accepted, etc.

Now I have a few notes about what will end up in the final 3.3 and
what I see are the biggest problems we face in that
but it's going to take a while to type up and I still need to finish
reviewing and revising the roadmap.

 I'm more curious as to suggestions as to what to accomplish in the 6
months after that.

As my crystal ball is in the shop, right now.

-- 
Dave Täht
SKYPE: davetaht
US Tel: 1-239-829-5608
http://www.bufferbloat.net