[OSM-dev] First post, &some questions :)

[Tom Hughes]

In message <be684e7e0707031530ja85a0ecvf574f8e5ae233729 at mail.gmail.com>
          Yann <yann.hamon at gmail.com> wrote:

> The www is swapping a LOT, which can usually be avoided with HTTP servers; I
> don't know lighttpd too much, if it was running Apache i would say you
> should lower down the max_processes value... the "applications" memory usage
> seems rather high, no?

I should perhaps explain a bit more how www is setup...

It is using lighthttpd, but that is not what does the real work, it
is just a frontend that farms out requests to a set of ruby fastcgi
daemons. Well actually it does serve a few static files (images,
javascript source, stylesheets etc) but all the real work is farmed
out to the fastcgi daemons.

There are currently six fastcgi daemons running - three are handling
the API requests (anything under /api basically) and the other three
are handling everything else (ie basically the web site).

The reason for the split is that the API daemons get quite big - most
of the time they will be around 4-500Mb each, so with three of them
on a machine with 2Gb RAM we can't reallyt have any more. Sometimes
one of them will get much bigger due to a particularly large query
and that is when things start to swap.

The daemons handling non-API requests are only a few tens of megabytes
in size and are much more manageable.

The reason the API daemons get so big is basically due to handling
requsts for map data and trackpoints, which for a busy area of the
map can be quite a lot of data.

Tom

-- 
Tom Hughes (tom at compton.nu)
http://www.compton.nu/