[Cerowrt-devel] shaperprobe tests

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

I am pleased to see mlabs has a server up for collections. Given that
there was other traffic on this link the results aren't too wildly
inaccurate.

I do find it amusing that it doesn't detect a shaper on any of my tests so far.

Recently I did a whole bunch of interesting traces with stuart
cheshire and you CAN detect a clear pattern for the inverse sqrt of
codel on a nearly idle link. I haven't looked at the guts of
shaperprobe since I saw it at gatech, but perhaps it's possible to
leverage it further in cero than has ever been done before and make it
more accurate. I sure hope that nobody's writing papers on the results
today...

I would be very interested on other people's results with and without
sqm enabled using shaperprobe. Current code and binaries for major
operating systems are at:

http://www.measurementlab.net/tools/shaperprobe

The one that is in cerowrt is 3 years out of date. Fixing that, but
for now running code from my x86 box:

I set my connection to 800 down and 220 up with nfq_codel.

d@k1:~/Downloads/shaperprobe$ ./prober
DiffProbe release. January 2012.
Shaper Detection Module.

Connected to server XX.YY.ZZ.QQ

Estimating capacity:
Upstream: 138 Kbps.
Downstream: 750 Kbps.

The measurement will last for about 2.5 minutes. Please wait.

Checking for traffic shapers:

Upstream: No shaper detected.
Median received rate: 112 Kbps.

Downstream: No shaper detected.
Median received rate: 746 Kbps.

For more information, visit: http://www.cc.gatech.edu/~partha/diffprobe

Then I did fq_codel for 25Mbit down, 4up


d@k1:~/Downloads/shaperprobe$ ./prober
DiffProbe release. January 2012.
Shaper Detection Module.

Connected to server XX.YY.ZZ.QQ

Estimating capacity:
Upstream: 3548 Kbps.
Downstream: 19125 Kbps.

The measurement will last for about 2.5 minutes. Please wait.

Checking for traffic shapers:

Upstream: No shaper detected.
Median received rate: 3472 Kbps.

Downstream: No shaper detected.
Median received rate: 18741 Kbps.

Then, for giggles, on an unshaped cable modem (which is roughly 25
down and 5 up with powerboost enabled, I think....)

Estimating capacity:
Upstream: 4794 Kbps.
Downstream: 96663 Kbps. # WTF?

The measurement will last for about 3.0 minutes. Please wait.

Checking for traffic shapers:

Upstream: No shaper detected.
Median received rate: 4666 Kbps.

Downstream: No shaper detected.
Median received rate: 28186 Kbps. # powerboost probably

It's choosing different servers for each of these tests so I would
expect that to be part of the issues, too.

-- 
Dave Täht

Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html