[Foundation-l] Freedom, standards, and file formats

[Gregory Maxwell]

On Mon, Sep 29, 2008 at 9:52 PM, Erik Moeller <erik at wikimedia.org> wrote:
> 2008/9/27 Gregory Maxwell <gmaxwell at gmail.com>:
>> The existing system correctly plays back for an overwhelming majority
>> of users (I posted some stats on this two years ago or so)
>
> Could you give a link, just for reference purposes?

Ugh archive searching.

Here is one:
http://lists.wikimedia.org/pipermail/foundation-l/2007-March/028210.html

"The last weeks data of 143,530 unique IPs shows that 79% of the IPs
hitting the Java audio player on toolserver have a Java Virtual
Machine. However, of those only 78% (61% of the total) have a JRE new
enough to use Jorbis. I believe Cortando has a resampler in it to
permit it work on the old MSFT JVMs, so it should do better.  The
number is skewed a bit by the fact that people without java are
unlikely to vist again soon since it didn't work, and are somewhat
better than my initial numbers, but I've also improved compatibility
dramatically since then."

(In the months that post the player was changed to something with
dramatically more JVM compatibility.)

We should basically expect the current success rate to be the total
number of people with a working Java install, which should include all
recent Macs, plus perhaps a percent with support from other supported
modes (Firefox 3.1alpha, Opera alpha, VLC, QT+XiphQT, totem plugin).
If it's not that high then there is something that we've done wrong.

Once Firefox 3.1 is released we should expect a non-trivial boost:
Firefox users upgraded to 2.0 very quickly, and firefox is a larger
share of Wikimedia's traffic than most of the published broswer
penetration numbers.

[snip]
> Things should improve significantly with Firefox 3.1, so we may want
> to hold off any new tests until it has gained some adoption.

It's always worthwhile to test, you can just discard the results if
they are not yet useful.  If nothing else some recent information
might help eliminate some Wikimedia induced lowness... It's only fair
to do measurements which reflect a little bit of "trying to make it
successful".