[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: percentage of popcon submitters



On Fri, 16 Jan 2009 08:45:12 +0100
Kjeldgaard Morten <mok@bioxray.au.dk> wrote:

> >
> > Thanks. Unless you setup some experimental method, any argument  
> > should reduce
> > to handwaving or extension of various particular examples..
> 
> Surely, it must be possible to get an estimate of the number of  
> downloads of important packages and security updates? I know these  
> downloads also are requested from mirror sites, but at least for the  
> official mirror sites their relative activity must be known?

How do you map the number of downloads to the number of users or
machines? I have dozens of chroots that I use for multiple reasons.
Now, maybe I should use an apt proxy but most of these are
cross-building chroots so that doesn't help as the proxy will have
amd64 packages and I need arm or armel etc.

Then you have the problem of people who maintain local mirrors (often
quite short lived ones).

It's just more handwaving - unless you want to count every chroot and
every local mirror (per architecture) as a separate "user".

There is no way of accurate counting unless access to the files is
restricted to a known number of download methods that all require user
intervention to proceed, at which point Debian would not be free.

The LinuxCounter method is completely arbitrary - the figures on the
site are guesswork and cannot be used in any other calculations.
LinuxCounter tries to extrapolate from 180,000 to 29,000,000 without
any real basis for such a leap of faith other than "we guessed 18
million some time ago and we have x% increase in our counter figures
since then, so increase 18 million by x%". I'm not knocking their
figures, just reiterating what is on the linux counter website -
reliable figures just do not exist and trying to create them usually
results in restricting the freedoms that attract users in the first
place.

http://counter.li.org/estimates.php

Linux Counter is no more or less reliable than popcon - both are wild
guesses from different perspectives. popcon is an wild underestimate,
counter could as easily be over as under. Nobody knows and in a very
real sense, nobody could ever know with any accuracy.

popcon is what we have, it is an indicator with known deficiencies that
always need to be taken into account when using popcon data as a factor
in any packaging decision but popcon, overall, is just more handwaving.
I find it amusing that we post popcon % figures to two decimal places
when the real error margins are completely unknown but it reflects the
size of Debian - if popcon didn't use decimal places, a vast number of
packages in widespread usage would have a popcon % of zero.

I could say that LinuxCounter is out by 30% or 70% or 150% and there
would be no reason to consider my guesses as more or less reliable than
the ones from LinuxCounter. The whole thing is a complete unknown.

-- 


Neil Williams
=============
http://www.data-freedom.org/
http://www.nosoftwarepatents.com/
http://www.linux.codehelp.co.uk/

Attachment: pgpARqXCOyHHB.pgp
Description: PGP signature


Reply to: