[OSM-Devserver] we need to reduce the server load!

Sven Geggus lists at fuchsschwanzdomain.de
Fr Mär 19 19:20:30 CET 2010


Hello altogether,

I'm going to write this in english because its important to anybody
using the devserver machine (gauss).

Currently we have at least two relevant projects (in terms of users)
which is the Reit- und Wanderkarte (TOPO Map ->
http://topo.geofabrik.de and the "All in One Garmin Map".

I think that in future the processing power needs to be scheduled in
a matter of project importance!

Sorry guys I would like to provide a seperate Machine to any one of
you, but I don't have them :(

The TOPO Map does currently have insufficient resources for tile
rendering. This needs to be changed!

Just as an example how important or unimportant the different
projects are, here are some access statistics (counted by quick and
dirty shell one-liners) from March 14th to March 19th:

Reit- und Wanderkarte: 2484
AIO Garmin Map:	1562
Navipowmmaps: 56
Gosmore Maps: 90

This is access from unique ip-addresses. Multiple downloads from one
ip-address are counted as one. Only real downloads are counted in case of
the Maps. 

So we have a real disproportion between processing power used and
overall importance of the single projects!  OSM2POWM is using rougly
the same processing power as the Garmin Map build process every day
while the latter has an importance which is about 25 times higher!

I also think that good Garmin maps are very important for the project
(AIO includes Mapper features like Openstreetbugs, Fixme etc.) and
the download numbers approve this.

This does not mean that I think of the others to be unimportant, but
I think it should be enough to produce up to date Navipowm and
Gosmore maps once a week.

The trouble is mainly with data conversion tasks and bulk database
imports that could be processed in a batch job like manner in future.

Probably its also a problem with the current lack of some kind of batch
processing scheduler (any hints for a suitable software?).

Currently all these jobs are started by cron when the users thought it
would be the best time to have them running.

Maybee the two important projects can also get migrated to the
machine which is currently not yet that busy (wms aka humboldt), but
I think this would only defer the problem rather than resolving it.

So what do you think?

Regards

Sven

-- 
Den Rechtsstaat macht aus, dass Unschuldige wieder frei kommen
(Wolfgang Schäuble)

/me is giggls at ircnet, http://sven.gegg.us/ on the Web