[Benchmarking] custom vs. stock builds of MapServer

[Frank Warmerdam]

Guys,

There was some question whether the custom build with optimization flags
that Paul did for MapServer might be substantially out performing "standard"
intel binaries.  To compare I installed stock FGS (http://www.maptools.org/fgs)
binaries (MapServer 5.4.2) on goliath and ran a test.  Since FGS seems to
lack bigtiff support, and fastcgi, I tried cgi - tiled tiff and got:

Label   Count   Avg     Min     Max     Errors  Throughput
1       100     201     95      593     0       4.9
10      200     684     110     6468    0       12.1
20      400     1262    110     17654   0       12.2
40      800     2659    103     43214   0       12.5

This compares similar to the benchmark MapServer 5.6 CGI results:

Label	Count	Avg	Min	Max	Errors	Throughput
1	100	362	97	3166	0	2.7
10	200	839	142	5208	0	11.0
20	400	1502	100	10515	0	11.4
40	800	2962	124	39417	0	11.3

(the single client case here may be giving poor results because I wasn't
always careful to prime the cache).

Anyways, the point is that we do not seem to be enjoying any noticable
boost due to the way mapserver was built for the benchmarking test,
and people should be able to achieve similar results with the stock
FGS MapServer builds for intel linux.

Best regards,
-- 
---------------------------------------+--------------------------------------
I set the clouds in motion - turn up   | Frank Warmerdam, warmerdam at pobox.com
light and sound - activate the windows | http://pobox.com/~warmerdam
and watch the world go round - Rush    | Geospatial Programmer for Rent