Thomas Corell wrote:
Timwi wrote:
I don't think the current Phase 3 software
can elegantly/efficiently
be enhanced to handle several languages within one server/installation.
I don't think a single server should handle several languages.
I suppose I wasn't being very precise. It doesn't have to be a single
server (which is why I said "installation" also). It should, however, be
one consistent system as a whole, not several independent systems. I'm
talking on a more ideological level ("what it looks like to the user")
rather than the technical implementation of it ("what it looks like to
the system administrator").
Take LiveJournal, for example. (Ignore for the moment that it's the only
website/codebase I know in real detail ;-) ) ... They have several
webservers and a load balancer that distributes incoming requests to
them. They also have several database servers, called "clusters", and
users' journals are randomly distributed over these clusters. Yet
LiveJournal is a single system in the sense that it is possible to add
another user to your own friends list (to name just an example)
seamlessly even when you're on different clusters. There is only one
additional "global" database (the "master") which stores all the
global
data (e.g. user preferences), which is tiny in size compared to the
clusters that store users' journal entries and the comments thereto.
What I find disturbing about the current Wikipedia setup doesn't really
have anything to do with running everything off multiple servers. It's
just that the different Wikipedia servers are too separate in the
following sense:
* You need an extra login on each of them.
* You have to change your preferences on all of them individually.
* You have to follow several watchlists/recent changes lists.
* You have to add Inter-Wiki links in *both* directions.
(Ideally, adding X to Y should add Y automatically not only to X, but
also to all other wikis that X links to.)
Ideally, Wikipedia should have one "master" database to take care of
passwords, preferences, watchlists, bans, etc. - which would be a
ridiculously small database compared to the language-specific contents,
whether or not it should also contain the entire User: namespace (which
would be an interesting topic for a separate discussion). Then each
language would have an own database; those databases could be put all on
one server first, and later they can be easily spread over several servers.
This is how I feel it /should/ work. Of course, if any of you can point
out disadvantages that I'm not seeing (except, of course, for the fact
that we don't have a software to do this yet), please feel free to reply
and discuss.
I don't know how come up with the matter, but I
think it's very useless.
Despite the fact that I can't make a lot of sense of this sentence, I
have a feeling it contains a trace of unnecessary polemics.
Greetings,
Timwi