I dont know enough about servers/linux as many of the other people here do
and sorry if I repeat any of the earlier advice because I havent read it.
Here's what has worked for our wiki although it has a smaller traffic
ranking than rationalwiki (which is about 22K right now or close).
Instead of Apache, use nGinx. Its good at serving files from what I've read
before. Use the simple file cache:
Also run APC. PHP type/mode is fpm-fcgi. Try it out and see what happens.
Some google searches for "apache vs nginx":
This has the same NGinx + php-FPM configuration I mentioned.
*-As you can see, nginx offers a higher transfer rate compared to Apache,
but also has less of a wait time between receiving the request and passing
a response back. -I also noticed that nginx can handle more requests per
second, and is able to ramp up as the load increases, however apache
remained pretty static on this front.*
In the 3rd and last comparison, you can see there are failed requests for
Apache but none for Nginx in that scenario. That graph is:
Could I ask how many page views a day (total for the site) you get when
this kind of crash happens? Or per hour.
On Fri, Feb 28, 2014 at 10:52 AM, David Gerard <dgerard(a)gmail.com> wrote:
On 18 February 2014 22:11, David Gerard
is getting hammered again. It
looks like MySQL is the
busiest portion - seriously just doing a lot of work.
Update: so we're finally taking action on this. We're just putting up
a new 4GB VM for Apache, we'll serve the main site from that which
should free up a ton o' RAM for MySQL on the increasingly loaded 8GB
main box. (We'll leave minor sites over there today, move them at our
leisure.) We have a bit of cash on hand (we have no paid staff) so can
put up a few extra boxes as needed. Thankfully, horizontal scaling is
PHP's strength. I have reminded Trent of the high quality
recommendations for other things to do to be found in this thread ...
(We just got sued by an aggrieved creationist. Thankfully WMF has
already set precedent for sec 230 applying, so we're not at all
worried about that. What we *are* worried about is well-wishers
knocking the server over again ... INCOMING!!!)
MediaWiki-l mailing list