[Benchmarking] Interested in searching for a bottleneck

[Andrea Aime]

Arnulf Christl ha scritto:
> Hey,
> I would be interested in looking into the current results a bit more 
> because it looks almost impossible to me that we should get so similar 
> results from two so different software packages.
> 
> My intuition tells me there is a bottleneck in the transmission between 
> the two boxes.

There is no transmission for shapefiles, yet the results are pretty
similar there too.
I run the tests with GS 1.7.0 lately and it was visibly faster
than both GS 2.0 and MapServer, meaning GS speed regressed since
last year.
I still haven't had the time to run a full set of tests to determine
where (thought I have a hunch: GS 2.0 was trying to compress
PNG files more and, silly me, I did not notice).

> Another reason could be that the overall processing time needed to 
> render the geometries is much smaller than the time required to retrieve 
> them - regardless of it being file access or database access. If only 5% 
> of the overall request is spent in the mapping software that would 
> explain the similarity of the results.

My observation was that the CPU's were used almost at full capacity when 
the thread count went up. A bottleneck in the network is usually paired
with low CPU usage instead.

> Maybe you already know this but what are the processor times split 
> between the map applications and the database / file access?

Unfortunately due to the way GS renders data I cannot tell. The
two things are mixed in a single loop to stream data out of
the server and render it instead of loading all up front
and rendering it later.

Cheers
Andrea

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