[158602] in North American Network Operators' Group
Re: Network Latency Measurements
daemon@ATHENA.MIT.EDU (Jim Gettys)
Wed Dec 5 10:17:53 2012
In-Reply-To: <74470498B659FA4687F0B0018C19A89C01A0F9031B9C@IL-MB01.marvell.com>
Date: Wed, 5 Dec 2012 10:16:41 -0500
From: Jim Gettys <jg@freedesktop.org>
To: Tal Mizrahi <talmi@marvell.com>
Cc: "nanog@nanog.org" <nanog@nanog.org>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On Tue, Dec 4, 2012 at 4:05 PM, Tal Mizrahi <talmi@marvell.com> wrote:
> Hi,
>
> We are looking for publicly available statistics of network latency
> measurements taken in large networks.
> For example, there is FCC's measurements (
> http://www.fcc.gov/measuring-broadband-america/2012/july
>
> ).
> However, we are looking for something more detailed that can show a large
> number of latency measurements taken periodically (preferably with as small
> a period as possible).
>
> Here are the datasets I'm aware of:
ICSI Netalyzr
The FCC measurements
MLabs http://www.measurementlab.net/
None of them, to my knowledge, take latency measurements "periodically".
I watched a nice demo at a talk about Mlabs a couple weeks ago of the
ability to query their data set and plot the results. They happened to
plot a million or so latency samples (and they have when those samples were
taken).
Just don't throw out the "can't possibly happen" outliers; bufferbloat is
bad enough (if you look at the Netalyzr scatterplots you can find here
http://gettys.wordpress.com/2010/12/06/whose-house-is-of-glasse-must-not-throw-stones-at-another/
you'll
see why...). Unfortunately, latencies measured in *seconds* are not only
possible, but not uncommon (e.g. my brother's DSL service has > 3 seconds
of buffering in each direction under load).
- Jim
