odd comment on a /. post

April 8, 2010

I normally don’t read slashdot. Mmcgrath pointed me to this from the article:

‘We have no phone home or registration process, so it’s always a guesstimate. But based on the same methodology that we came up with for the 2008 number, our present belief is that it’s somewhere north of 12 million users at the moment,” Chris Kenyon, vice president for OEM at Canonical, told InternetNews.com.’ Just in case you were wondering, Fedora still claims more — actually almost double, at 24 million.”

I was confused by this – since fedora has our numbers posted. So, when I read them it looks like in a 20-week sample period we found a total of 2million unique ips.

And that over the lifetimes of f7,f8,f9,f10,f11 and f12 we’ve tracked 24 million connections from 20 million unique ips.

We never claimed 24 million users – just 24 million connections from 20 million unique ips.

In the last 20 weeks we claimed 2 million unique ips

At most we might claim 2 million user connections in the last 20 weeks, but apparently the /. “editors” cannot read the tables.

oh well.


9 Responses to “odd comment on a /. post”

  1. Jag Says:

    And this is different how? I remember when they claimed the total number of bugs in bugzilla was the bugcount for a (then) just released version of RHL.

  2. skvidal Says:

    yes it’s not just /. but also apparently some random guy at linuxplanet.com, the source of the article.

  3. Ryan Rix Says:

    Slashdot at this point is nothing more than a funny joke to me. Their editors are ridiculous, their articles are sensational, and their community is a bunch of complete morons.

  4. jef spaleta Says:

    Now if only the quoted Canonical exec would publish the methodology he referred to we could see how it stacks up to how we count in Fedora. For all we know the article writer’s weird interpretation of our Fedora unique ip collection methodology is actually comparable to what Canonical is doing to get their number.

    In point of fact, the only published stats on Ubuntu I can find are popcon and I’m pretty confident that the Ubuntu popcon statistics are being run in accumulation mode across all releases of Ubuntu. This differs from how Debian operates popcon, which has a 20 day activity window so that any UUID not reporting in during the last 20 days gets dropped from the stats.

    I’ve looked really closely at Ubuntu and Debian popcon data and nearly 40% of the UUIDs being counted in Ubuntu’s popcon are from EOL’d versions of Ubuntu. On top of that all the arches so monotonic growth in UUID’s reporting in..including sparc and ppc. That’s confirmation for me that Ubuntu is doing accumulation UUID stats instead of snapshoting current installbase activity like Debian does.

    If Canonical is using popcon stats with some fudge factor as a starting point to pin their estimate… then its effectively comparable to the article’s 24 million count for Fedora…and both would be bogus as a measure for current userbase size.


  5. Paul Frields Says:

    The fellow posting at LinuxPlanet isn’t even a totally random guy. He’s a journalist we regularly work with, and I’ve discussed our statistics with him before, including how we gather them and what we measure. Two things that do seem completely random in that article, though:

    * Taking our statistics and methodology out of context as a measure of total user count

    * Taking some Canonical employee’s word for their user count without any methodology to back it up

    • quaid Says:

      FWIW, I’ve found the statistics confusing the way they are written. It might be good if we draw some reasonable conclusions from the tables and put them on the page. These are conclusions that are obvious to the more mathematically minded of us, which are more obfuscated to those of us with the other mind.

      I’ve hesitated at writing such conclusions primarily because I’m too much of the latter and not enough of the former.

      • skvidal Says:

        I’ve got no objection to making it a bit more clear – but I think what the page we have does tell you is: DO NOT DRAW GENERAL CONCLUSIONS BASED ON THESE NUMBERS.

      • jef spaleta Says:

        There is an additional metric that we currently produce but we don’t highlight. In the map generation we use a week’s worth of unique IPs from MirrorManager to create the heat maps. We could take that “weekly active” unique ip address count, generated daily, and make it more prominent on the stats page instead of buried on each and every map.

        What’s great about the “weekly active” count is stale ip addresses roll out and new ip addresses roll in and that prevents the accumulation of stale unique ip addresses to a large degree.

        It doesn’t prevent all duplicate counting from roaming clients…nor does it solve the problem of under counting because of shared ip addresses.
        but it does prevent misinterpretation that allows ballooning of the type we see here.

        Example.. right now some of “weekly active” counts from the maps are:
        F12 32bit: 120124
        F11 32bit: 53712
        F10 32bit: 33138



  6. skvidal Says:

    Paul, thanks. It just shows how little I know and how little research I did about him.

    Hey! Maybe I should be a writer for linuxplanet. 🙂

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: