[Tilecache] Number of simultaneous hit for tilecache (on apache with mod_python)

[Christopher Schmidt]

On Fri, Sep 05, 2008 at 03:36:13PM -0400, Martin Ouellet wrote:
> Hi,
> 
> I've made a cache of 30k tiles on disk (200Mo).  The network team that I
> work with want to test tilecache to see how he will behave if our Web site
> receive a huge amount of user simultaneously.  So I send them a list of URL
> with ramdom BBOX so they can call tilecache in a loop to test it.
> 
> They simulate 8 users at the same time and the result is pretty good
> actually (about 1300 tile /sec.).  but to obtain this result, the CPU is
> only running at 27% of his capacity and even if they increase the number of
> user, this percentage remain almost the same.  So we were wondering how we
> could use more efficiently the CPU and at the same time, increase even more
> the performance of the tile serving
> 
> Does someone have an explication for this situation?
> Can we suspect an apache, python or mod_python, windows limitation?

1300 tiles/second is 4 times higher than any number I've ever tested
with. If you really want to, you can just pre-generate your whole cache,
and (if you're using OpenLayers) use a Layer.TileCache, which directly
accesses a web-accessible cache. That way, you skip TileCache entirely.

My first guess of limitations is "network", my second guess is "apache
children". Looking at time per request in some way may help identify if
requests are uniformly slow (indicating a machinewide limitation) or are
mostly fast with a few slow (which might mean a resource limitation;
memory low, running into swap, etc.)...

If your application actually requires 1300 tiles *per second* though,
you should be aware that I don't trust TileCache to that level at all :)

Regards,
-- 
Christopher Schmidt
MetaCarta