RRUL test prototypes

RRUL test prototypes

[Dave Taht]

Toke made enormous progress on a RRUL test prototype over the last 2
months. It requires a recent netperf, python, python's matplotlib, and
netperf-wrappers.

For basic install on a debian/ubuntu box:

sudo apt-get remove netperf
svn co http://www.netperf.org/svn/netperf2/trunk/ netperf-2.7
cd netperf-2.7
./configure --enable-demo
make
sudo make install
cd ..

sudo apt-get install python-matplotlib
git clone git://github.com/tohojo/netperf-wrapper.git
cd netperf-wrapper
./netperf-wrapper -o mytest.ps -p all_scaled -H snapon.lab.bufferbloat.net rrul

This will also output a timestamped datafile that can be replotted with (for
example) the icmp_cdf plot type. I tend to use that to feed multiple
test runs into a cdf plot, for example...

/netperf-wrapper -i 'rrul-2012-12-22T18:48:00.965185.json.gz' -i
'rrul-2012-12-22T18:49:49.484438.json.gz' -o icmp_cdf.ps -p icmp_cdf
rrul

It'll get simpler as time goes by... (also supported are .svg and .png
output formats)

We have a variety of other tests in the netperf-wrapper tests subdir.
6vs4competing,
rtt_fair, simple tcp bidirectional etc, etc. See also the README.rst
for more details.

You can also run mtr at the same time in another window to see where
the bottlenecks are. If you are not based in the US, you should setup
a netperf netserver nearer by than
snapon (which is in california) - there's one in germany for example.

I have managed to build all this stuff successfully on the mac, but
packaging it up is a bear.  Will be working on a simple apt-get
package for ubuntu soon, fedora and arch, too.

Still needed are a decent voip and gaming test. I have a pretty good
mental picture of what most gaming traffic looks like now...

Re: RRUL test prototypes

[Rick Jones]

On 12/31/2012 03:01 PM, Dave Taht wrote:
> Toke made enormous progress on a RRUL test prototype over the last 2
> months. It requires a recent netperf, python, python's matplotlib, and
> netperf-wrappers.
>
> For basic install on a debian/ubuntu box:
>
> sudo apt-get remove netperf
> svn co http://www.netperf.org/svn/netperf2/trunk/ netperf-2.7
> cd netperf-2.7

strictly speaking I have not declared a netperf-2.7 and at least as I'm 
typing am not sure if the current top-of-trunk will become 2.7.0 or 
2.6.1 (I have to see if I've changed the control messages size since 2.6.0)

BTW, in Ubuntu at least I have gotten into the habit of not apt-get 
removing the supplied netperf, but sudo service netperf stop instead, 
then build my bits and then copy the netperf and netserver binary to 
/usr/bin (where Ubuntu puts them) and an sudo service netperf start. 
That way I do not have to remember to start a netserver each time I 
reboot the OS.

>...
>
> You can also run mtr at the same time in another window to see where
> the bottlenecks are. If you are not based in the US, you should setup
> a netperf netserver nearer by than
> snapon (which is in california) - there's one in germany for example.

I'll mention again just for completeness that a top-of-trunk 
netperf/netserver includes support for a required passphrase to be 
provided by netperf to netserver to run a test between them.  It is 
somewhat lame as the passphrase is passed in cleartext, but probably 
still better than a wide-open netserver on the Internet.

happy benchmarking,

rick