Re: [Cerowrt-devel] cerowrt 3.3.8-17 is released

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

Yes. As it is highly important to me to only get bug reports like this one

http://www.bufferbloat.net/issues/330

only every 5 years or so, surviving stress testing is paramount.

to whit, I'm planning on dropping bind from the next release,
switching to dnsmasq and busybox ntp, and disabling or dropping the
underused polipo proxy -

This would free up somewhere around 8-12MB of memory on boot.

in addition, eric dumazet has proposed two patches on the codel
mailing list that should reduce skb (packet buffer) size when needed
and possible.

https://lists.bufferbloat.net/pipermail/codel/2012-August/000422.html

I'm also looking into a possible memory leak on an error path...

And:

As noted in a prior message there are also some improvements to codel
that could be made, particularly to have a shared buffer cache across
multiple hardware queues.

I would like to be able to thoroughly stress test codel and fq_codel
in the next release of cerowrt.

on 128MB systems such as the 3800 I would hope that we'd have enough
memory for things like bind, but as there are also 32MB systems in the
openwrt mix, doing what we can, throughout the stack, to not run out,
is a good thing.






On Wed, Aug 22, 2012 at 11:23 AM,  <dpreed@reed.com> wrote:
> Ahhh,  thanks, Sebastian.  Now I understand what you are doing is a stress
> test.


-- 
Dave Täht
http://www.bufferbloat.net/projects/cerowrt/wiki - "3.3.8-17 is out
with fq_codel!"