With more and more images and large media files on the
rise, we could
noticeably reduce the server load by distributing these files.
How? Well, first we'd need some (5 to 10?) mirrors willing to sync their
data on a daily basis.
I thought we had discussed this before. It was decided that it would
make way more sense to use a ready-made solution (Akamai) than to make
one's own, but it was also decided that a change to the software would
first be necessary. Currently, a tag like this:
produces an <img> tag directly from just the filename:
where the "e" and the "2" depend only on the filename.
The problem with this is that these paths do not change if the image
changes. This means that overriding an image by uploading a new one to
Wikipedia would still serve the old version of the image from Akamai
until Akamai is updated.
Hence, a change to the source code is necessary that makes the file path
dependent not just on the filename, but also on either the file itself
or the time it was uploaded; however, this increases the load on the
database because it means an extra query for every pageview that
contains an image.