[Cerowrt-devel] using other packages | also, hi!
dave.taht at gmail.com
Fri Apr 6 15:49:27 EDT 2012
On Thu, Apr 5, 2012 at 2:43 AM, Maciej Soltysiak <maciej at soltysiak.com> wrote:
> I am new here, just subscribed. First of all thanks for all great work done
> on bufferbloat and incorporating that into wonderful cerowrt.
> Can anyone please tell me which Open WRT package set is Cero WRT based on?
> Is it
> a) openwrt trunk -
> b) openwrt backfire -
> c) other?
Other. It is based on trunk + a few extra packages and kernel options.
I use git from nbd's repository rather than svn.
I believe in a previous message the correct procedure for pulling was
This has to get cleaned up and documented better but I only got back
to working builds on monday...
> Also, if I want to add a package, say minidlna or ntfs-3g drivers from
> openwrt, which packages should I use?
Presently I am pulling directly from the openwrt packages repo. I will
freeze fairly soon and try to maintain a stable packages repo on
I note that I found the performance of ntfs-3g to be dismal and you
are far better off to reformat said drive for ext4, if you intend to
use openwrt/cerowrt as a nas. (far better off as in , 50x better off)
> Maciej Soltysiak
> p.s. I have observed very real improvments some time ago already by loading
> a custom TomatoUSB w/ bufferfloat mods (kernel driver buffers, txqueuelen,
> ring buffers, ECN, etc) to WRT54GL. Now I am using WNDR3800 with latest
> Cerowrt (just flashed it yesterday for the first time) and this also worked
> well towards decreasing bufferbloat, thanks! (measured by netalyzr)
The principal thing that cerowrt does out of the box different are BQL
+ SFQRED on ALL interfaces (not just the external interfaces).
This introduces a degree of statistical noise in the streams that
helps a lot in the modern era.
Please feel free to publish your preliminary results here or on the
wiki. It is still very early days, however, and the default AQM for
bandwidth shaping remains in flux and is off by defautl (and buggy
A new one (debloat) - is in progress, and is on by default, and not
buggy when on.
Two tools exist in the repository and on the router to help -
shaperprobe (prober) can measure bandwidth outgoing to some degree, as
The existing debloat script can be fiddled with with decent results
between 4-20Mbit uplinks by merely running this:
IFACE=ge00 UPLINK=whatever_in_kbit QMODEL=htb_sfq_red /usr/sbin/debloat
(determine whatever_in_kbit from any of various tools like the above
debloat on an interface can be disabled with:
tc qdisc del dev ge00 root
I would LOVE some netanlyzer results from that from people with adsl,
cable modems, and other forms of uplinks, with debloat as it stands
on, and off. (and as I noted, sfqred is ON on the internal interfaces
I am aware there are issues with ADSL, as well as a persistent
underestimation of overall bandwidth in shaperprobe that is in the 20%
range at 10Mbit, but fairly close at 2. I found when I last played
with netanyzler that it did a much more accurate job when connected
behind the router wired.
There are numerous tweakable params in the debloat script, it's in
lua, starting from the bottom up or top down it should be easy to
modify further (notably for adsl)
> Cerowrt-devel mailing list
> Cerowrt-devel at lists.bufferbloat.net
US Tel: 1-239-829-5608
More information about the Cerowrt-devel