[ubuntu-web] Question about MoinMoin wiki scalability

Attila-Mihaly Balazs dify.ltd at gmail.com
Fri Apr 20 18:48:06 UTC 2012

Hello all!

I'm trying to scale a MoinMoin based wiki and had hoped that somebody
could give me a couple of pointers since wiki.ubuntu.com is running
the same software. What I have currently:

- a Linode VPS with 2G of RAM, 4 cores and Ubuntu 11.10 64 server (of
course :-))
- Nginx, uWSGI and Python 2.7.2
- MoinMoin 1.9.4
- Requests per day: ~200k

That is a light load in average, however at the peak times I still get
gateway timeouts (around 6% of the requests result in timeouts - I
have a timeout of 4 seconds set in Nginx). What I did until now:

- made sure the disk has dir_index and noatime, nodiratime
- set up nginx to serve the static files (related to the skin for
example) directly
- reduced the keepalive timeout from 65 to 30 seconds

According to blitz.io this setup maxes out around 100 reqs/sec but
this just doesn't seem enough. Also, the setup seems to be very much
CPU bound. Any ideas on how to scale it? My next idea would be to use
NFS for sharing the wiki between two computers and use a second set of
CPUs, however I didn't find anybody documenting such a setup and I'm
somewhat weary of NFS. I would really appreciate if somebody could
give a couple of pointers about the wiki.ubuntu.com setup.

Attila Balazs

PS. I apologize if this mailing isn't proper for this kind of
discussions and would appreciate it if somebody could point me to the
right one in this case.

More information about the Ubuntu-website mailing list