Debloat testing2.6.39-rc4 kernel...

[Jim Gettys <jg@freedesktop.org>]

I've built (and am running) John Linville's latest debloat-testing 
2.6.39-rc4 kernel. It seems to behave alot better than the 2.6.38 kernel 
did for me on the Intel iwlagn driver, given (all of) about 10 minutes 
of testing.  We haven't tried to play with SFB yet; we've mostly been 
router hacking.

I put the packages Dave and I are running (64bit Ubuntu 10.10) on 
http://mirrors.bufferbloat.net/Builds/

I also did a quick script for setting the wireless txqueuelen to 
something more reasonable at driver up time; you can find it at: 
http://www.bufferbloat.net/projects/bloat/wiki/Wlan_script; at the 
moment, it's set to only 32 packets and my latency under load is staying 
somewhat reasonable with at least some decent throughput.  But given 10 
minutes of smoke testing... Dead puppies are possible.  No warranties, 
express or implied...

I documented this in: 
http://www.bufferbloat.net/projects/bloat/wiki/Linux_Tips

If people have the ability to set up and Ubuntu PPA, or can run packages 
through RH's build system (Koji, iirc), that would be good, so that 
people who want to test don't necessarily have to build from source all 
the time.  It would be really good to get debloat testing kernel 
packages built automatically on an ongoing basis from John's tree,one 
way or the other.  The other way to automate this is by using the server 
we have now that can build kernels in finite time; we can set up VM's on 
to build them. But that is a lot more on-going build system maintenance

So I'm a fan of using the upstream distro build systems for this if we 
can; we have lots of build work to do for the work that is underway to 
debloat the WNDR3700's running OpenWRT.  Volunteers more than welcome 
for either/both.
                     - Jim