[Benchmarking] MapServer test run bringing the server to its knees?

[Andrea Aime]

Hi,
I was looking into the long time it takes for MapServer run
to perform the full test suite (5 hours and counting now) and noticed
the server is almost at its wits end memory wise, the
fastcgi mapserver processes are eating all the memory away:

[aaime at goliath ~]$ ps aux | grep mapserv560beta3.fcgi
apache   10505  6.8 11.3 1069428 943992 ?      R    03:38  20:03 
/var/www/cgi-bin/mapserv560beta3.fcgi
apache   10514  6.8 11.2 1060832 935224 ?      R    03:38  19:55 
/var/www/cgi-bin/mapserv560beta3.fcgi
apache   10515  6.8 10.9 1038032 912096 ?      S    03:38  20:01 
/var/www/cgi-bin/mapserv560beta3.fcgi
apache   10516  6.8 11.5 1088292 963920 ?      R    03:38  19:58 
/var/www/cgi-bin/mapserv560beta3.fcgi
apache   10517  6.8 11.3 1070548 945036 ?      S    03:38  19:57 
/var/www/cgi-bin/mapserv560beta3.fcgi
apache   10518  6.8 11.2 1059140 933832 ?      R    03:38  20:04 
/var/www/cgi-bin/mapserv560beta3.fcgi
apache   10519  6.8 11.6 1091516 965800 ?      R    03:38  19:58 
/var/www/cgi-bin/mapserv560beta3.fcgi
apache   10520  6.8 11.1 1054516 929068 ?      R    03:38  20:02 
/var/www/cgi-bin/mapserv560beta3.fcgi

That is, in summary, each fastcgi process eat around 900MB of resident
memory. Looking at the machine free memory in MB:

[aaime at goliath ~]$ free -m
              total       used       free     shared    buffers     cached
Mem:          8118       7626        492          0          6        217
-/+ buffers/cache:       7402        716
Swap:         5535        557       4978

The actual free memory, excluding FS buffers, is 700MB (out of 8GB total).

I guess this is affecting negatively the recorded performance as
well, at least for shapefile tests, since the shapefile does not
manage to get cached fully.

Cheers
Andrea


-- 
Andrea Aime
OpenGeo - http://opengeo.org
Expert service straight from the developers.