Here is what I originally wrote about performance testing the Jython-based TileCache:
http://whitney.textdriven.com/blog/2007/06/20/testing-out-jon-blowers-jython-powered-tilecache/
My "performance testing" was basically a quick look at the requests / time statistics given by Tomcat for both Geoserver and TileCache after playing with the OpenLayers front-end for a few minutes.
After fighting a few problems, I have Jon Blower's Jython TileCache running again. I also have a preliminary version of my tile cache, JTileCache, running with JCS. I did a quick test:
-- Tomcat with 256 MB allocated
-- Each test is a fresh restart of Tomcat, followed by a first run (to populate any cache), then a few following runs to establish a consistent time
-- The test is to fetch the same 20 sort-of-random image tiles from topp:states (in jpeg)
-- Time is established based on 'time curl...', so probably only useful for relative time comparisions
Geoserver:
1st run: 1.964s
Consecutive runs: ~1.1s
Jon Blower's Jython TileCache (disk cache cleared on first run):
1st run: 4.450s
Consecutive runs: ~0.25s
JTileCache (with LRU memory cache only):
1st run: 3.685s
Consecutive runs: ~0.41s
So, assuming the test overhead is minimal:
Geoserver: ~0.9 requests / sec
Jython TileCache: ~4 requests / sec
JTileCache: ~2.4 requests / sec
[ Looks like I have some work to do ... and next time I should actually test before jumping to conclusions
]
I think it would be useful to repeat these tests with a more complicated map and a better tile-fetcher (aka cache seeder).
-- Chris Whitney
On Aug 2, 2007, at 7:01 AM, Andrea Aime wrote:
Jon Blower ha scritto:
Hi Andrea,
I didn't do any benchmarking at all and to be honest I don't use it
myself (having developed a faster solution for my particular problem).
However, this is really useful information. Have you published these
results anywhere? If not, would you consider adding them to my
JTileCache Wiki page (I can give you a login)?
I wonder if we can achieve better performance in JTileCache by
compiling the Python code to Java classes instead of running it
interpreted through Jython...
Hey Chris,
anything to add about this? I just reported what I saw in a mail.
Cheers
Andrea