[fab] Metrics: What we *could* get

[Josh Boyer]

On Fri, 2006-10-06 at 14:14 -0400, Greg DeKoenigsberg wrote:
> On Fri, 6 Oct 2006, Christopher Blizzard wrote:
> 
> > There's another interesting aspect of having something on the machine 
> > that allows someone to be counted.  It follows the software all the way 
> > from where we put it up, through the mirrors and down to the 
> > installation.  It's much more reliable about reporting something in that 
> > sense, rather than collecting stats from mirrors and download sites.
> > 
> > It also reports the act - "I'm using it" instead of "I downloaded it."
> > 
> > And last, in some ways it's a lot less nefarious than tracking 
> > downloads.  It's a result of a positive act on part of the user, as 
> > opposed to tracking on the backend without any kind of positive assent 
> > from the user.
> > 
> > --Chris
> 
> Yep.
> 
> If we're going to bother doing actual engineering work for a measurement 
> solution in FC7, then it should be:
> 
>   a. Voluntary;
>   b. Anonymous;
>   c. Incredibly comprehensive;
>   d. Clearly beneficial to users.
> 
> And when I say "incredibly comprehensive," I mean that it should collect 
> hardware information and package information, and we should focus on 
> deriving *useful* data from these metrics.  Like:
> 
>   * What hardware works well, and what doesn't?
>   * What software packages are widely installed?
>   * What's the reliability of a package, based on a ratio of
>     "open bugs" to "instances installed"?
> 

http://klive.cpushare.com

Package it for Extras.  It'll even tell you what version of the kernel
folks are running.  Modify it a bit if you want it to gather more
statistics and have it report to your own server.

josh