[Opensim-users] Prims and CPU usage?

[John Sheridan]

Gang,
    After trying a few of the higher prim oar files I've seen fly 
through here I've begun to notice a rather curious trend, although I'm 
unsure if its to do with my configuration or if others have seen this as 
well.  After trying Nebadon's Art City oar on a local standalone (Mono 
2.0.1 / SQLite, Basic Physics on Linux Ubuntu) I found that after 
everything loaded and committed to the database (which on its own took a 
good hour and a half probably due to SQLite) upon restarting OpenSim cpu 
use for the OpenSim process was pinned at 30% when idle. 

  Oddly enough, since I started building my Ravenport sims (Mono 2.0.1 / 
MySql, ODE / ZeroMesh on Ubuntu) I've as well noticed a steady increase 
in cpu use as the build progressed, however until now I thought this was 
due to scripts or possibly a few rogue physical prims that may have sunk 
under ground or flown off the sim edge.  Currently when idle the 
Ravenport server is using 25% of the cpu at ~13,000 prims where a month 
ago it bordered around 10 to 15% at roughly 7,000 prims.  Out of 
curiosity I saved the sims out to an oar and began dissecting them on 
the same local configuration I tested the Art City  oar on.  
Surprisingly after removing all scripted objects there was no change in 
cpu use - it was still at 25%.  It wasn't until I began deleting the 
buildings that I began to notice a change where as the sims emptied out 
the cpu use decreased down to between 1 and 2% once all prims were 
deleted. 

  To verify, I decided to try again this time with a clean OpenSim 
instance of the same config where I had set a scripted prim to create a 
grid of default .5x.5x.5 cubes.  The first round was with a grid of 
13,000 prims which upon persisting and rebooting pinned the cpu at 30%, 
the second test I added another 13,000 which brought it up to 40%.  
After adding another 13,000 (39,000 total) the cpu was pinned at a solid 
100% after a fresh restart. 

  I have to admit, I do find this to be quite baffling.  In all I'd 
expect the sim to use more memory with a higher prim count but the cpu 
should remain about the same, especially given that my test environment  
was using basic physics where all prims were rendered as phantom. 

  The oar for the 39,000 prim sim is located at 
http://www.pseudospace.net/primzilla.oar if anyone else would like to 
test it on their own configuration.

Thanks, :)

 - John / Orion Pseudo