[BlueOnyx:12199] Re: Hey Michael! Any idea how many BO boxes are out there?

[Michael Stauber]

Hi Roy,

> Possibly the easiest question of the day.
> 
> Im working through an issue with a skeptical developer who has never 
> heard of the platform before and I am curious. Does anyone have any idea 
> approximately how many BlueOnyx boxes are out there in production around 
> the world?
> 
> I assume Michael has an idea approximately how many due to the yum 
> update connections in the log files. (then again I could be completely 
> off base making wild assumptions)
> 
> Right now my gut says "alot" but I need a wee bit more accurate number 
> than that to illustrate that this isnt some half-baked, home grown 
> distro thats being used by one guy and his buddies ;)

Check the bottom of this page:

http://www.blueonyx.it/index.php?page=features

BlueOnyx server tally:
----------------------
BlueOnyx is running on more than 23026 (5106R: 17914 + 5107R: 2243 +
5108R: 2159) servers according to our internal tally, which is based on
automated YUM update usage counts.

Also check the Devel-Blog article "Apache, the 500 pound gorilla" at
this URL, which explains our mirror situation a bit and also has a map
and a table of our YUM update statistics:

http://www.blueonyx.it/index.php?mact=News,cntnt01,detail,0&cntnt01articleid=155&cntnt01origid=62&cntnt01pagelimit=100000&cntnt01returnid=62

I wouldn't say that our server tally is 100% accurate, because due to
the way that the data is generated there are some unknown factors.

It works like this:

Whenever a BlueOnyx does a "yum update", then it fetches the list of
mirror servers by calling a certain PHP script on updates.blueonyx.it:

mirrorlist=http://updates.blueonyx.it/mirror.php?release=$releasever&arch=$basearch

$releasever is the version of the OS. Like 5.9 for CentOS 5.9 or 6.3 for
CentOS or Scientific Linux 6.3. Additionally $basearch is either i386
for 5106R or 5107R or x86_64 for 5108R.

The combination of the version number of the OS and the basearch allows
us to determine if the access is from a 5106R, 5107R or 5108R.

Whenever mirror.php is called, we make an API call to our Piwik install
on https://stats.blueonyx.it/ to record the transaction. Now usually
Piwik allows very detailed statistics if you integrate it into a
webpage. But as we use the API and the the BlueOnyx servers access it
via YUM, we can't set a cookie from Piwik that would allow a much better
tracking of recurring visits. So our Piwik API calls are logged only if
the user agent correspondents with YUM and as user agent we manually set
the BlueOnyx platform type (5106R, 5107R or 5108R). Additionally we log
the IP and architecture together with a timestamp into a separate MySQL
database that's independent from Piwik.

When all is said and done, mirror.php will spit out a list of currently
active BlueOnyx YUM repositories, from which your YUM then picks the one
that reacts the fastest.

You see where the weakness is: BlueOnyx boxes on dynamic IP's might get
counted repeatedly - each time from every IP they happened to use over
time. Or if several BlueOnyx boxes are in an internal network, which
communicates to the Internet via a single IP, then those multiple boxes
will get counted only once.

Likewise: Boxes that never use "yum update" or which are hard wired to a
specific YUM repository will never be counted.

These statistics didn't exist from day one, though. I started them in
2009. From the discussions on the developer list I grabbed these
historical milestones in statistics:

24.09.2009: 1003
29.11.2009: 1977
10.04.2010: 4091
08.08.2011: 15.012 (5106R: 14931 - 5107R: 76)
05.10.2011: 16.003 (5106R: 15308 - 5107R: 405  - 5108R:	79)
31.01.2012: 17.779 (5106R: 16049 - 5107R: 797  - 5108R:	370)
15.02.2013: 23.026 (5106R: 17914 - 5107R: 2243 - 5108R: 2159)

AFAIK we broke 20.000 around sometime in summer of 2012, but I didn't
record the exact date.

-- 
With best regards

Michael Stauber