[GRLUG] Letter from Comcast

[Adam Tauno Williams]

On Tue, 2013-04-23 at 10:40 -0400, Brad DeVries wrote:
> Ben, I don't have an upper tier connection here in Hudsonville,
> although I feel like it because I pay almost $60/mo, and I just ran 5
> speedtests with the old DOCSIS 2 modem with vastly different results:
> I'm not sure how accurate these test are

These tests are approaching whichever axis is the 'meaningless' one.

But I do generally take someone making a big deal out of a web-site
'speed test' as rather solid evidence that they do not know what they
are talking about.  Once 'speed test' is mentioned one can comfortably
return to one's beverage, you aren't going to miss anything.

Meaningful performance metrics are hard, tedious, and full of caveats.
Arbitrary, meaningless, and off-the-cuff metrics are easy to do ... but
do little other than dilute the conversation and promote pomposity.

A meaningful metric must at least be derived from a long-duration
sampling as IP throughput tends to be quite wobbly - this is something a
web browser cannot accomplish.  And to test 'the last mile' the remote
node needs to be very very extremely close to, if not at, the other end
of your 'last mile'.  

Gathering SNMP stats for your edge router is probably much more
meaningful - what throughput does the interface actually see?  Does it
hit a ceiling?

And Define "speed" - throughput or latency?  And throughput to/from
where for what traffic pattern?  Latency to/from where for what traffic
pattern?  "speed" is a bogus layman's concept and itself is just not a
technically useful term.

If you're comfortable with your network's performance - then it has
enough 'speed'.