On Fri, 18 Mar 2005 10:30:52 +0200, Domas Mituzas
<Domas.Mituzas(a)microlink.lt> wrote:
Scalability happens, when performance grows in linear
mode with hardware
and even uses some synergy. This is one of mediawiki performance development
goals. Make it distributed as much as possible. Make it as async as possible.
I think the big lynchpin right now is the fact that we have one master
database. I don't know how we get from here to dual (or more) master
databases. I think we also have a disk deficiency. Things like PDU's
are easy to fix, and I think jwales/chaper are already on that.
Taming cheap horsepower is a challenge, but it helps
project to work on low
The last installation I worked had thousands (many thousands) of dual
1.3ghz p3's, and we did just fine. Taming cheap horsepower is a
challenge, but not an insurmountable one. It takes infrastructure and
a little bit of work.
ones :), and who know meanings of 9's after comma.
Sure, you've got possibility
to brainstorm and bring all kinds of ideas, but we aware, that those won't be
accepted as absolute truth. People do not need prophets when critical thought is
involved :) Though, any contribution to development is here highly appreciated.
Things I say are clearly not being accepted as absolute truth. I
expected as much. I did not expect the personal attacks and
name-calling.
What I most want to address is these things:
* Why are we having outages?
-- When I say outage, I mean people popping up in #mediawiki or
#wikipedia and saying "wikislow" or "wikidown". It seems that we
actually have a system that is working for now, but there are a few
magic bullets that seem to sneak into the system from time to time and
cause utter havoc for the developers, or if you want to be honest
about it, the SA's and DBA's.
* What can we do to streamline or otherwise improve the performance of
the project?
-- The "fancruft" thing has been nagging at me since David started
attacking me on it. I really don't care what kind of article it is,
but would it be possible to create "clusters" of articles (represented
in the database as such) to make it easier for the database engine to
find what it wants? I'll continue these thoughts in reply to David's
message.
alex
--
Alex Avriette
avriette(a)gmail.com