[Opensim-dev] 20,256 prim test

[Teravus Ovares]

Greetings everyone

I was having fun tiling the ground and above with .25 meter cubes and ended
up with this;
http://opensimulator.org/images/d/d0/My20000Test_001.jpg

This test was conducted on a 100MB ethernet network.

The instance is running 3 regions, two with no prim in it.
The sim is running ODE with a single agent.

 The client is running on a different machine then the sim.
This was using the mySQL DB interface in Grid Mode

at 16,750 prim, the instance of OpenSim was using up 45% of a dual core 3Ghz
processor(one core)
at 20,256 prim, the instance of OpenSim was using up 22% of a dual core 3Ghz
processor(both cores)
In both cases the memory usage was around 145MB of RAM
using 58 Threads
758 handles
avg 1600 context switches

.NET Perf Info
11,750,740 = # Marshaling with unmanaged code(ODE)
2,839 = # GC handles
0.10 - 0.36  = % Time in GC
4,840 = Classes loaded
69 = Total Assemblies

.NET Locking and Threads
2 = locking contentions /sec
75,505+ and increasing = Queue Length peak
58 = Logical threads
4,294,967,033 = physical Threads
322 = current recognized threads


A couple of notes.

It's really difficult to get to 20,000 prim.    The reason is DataStore
Database backup takes an hour (based on prim duplications).
At least with mySQL, this can be done asynchronously, I suggest we do that,
otherwise you can DOS a sim simply by copying prim.
IMHO, this is the *WORST* performance bog that we've got.  Even worse then
the packet object create and discard.

When logging into the sim, it takes about 3 minutes on a 100MB network to
get all 20,256 prim at 600Kb/sec.  During that time, you can't really move
because everything is being pushed through the Task throttle (this is by
design).   Before this becomes general practice, we'll need to make sure we
limit the amount of prim that go out at a time to a client.

One 3Ghz core could handle approx 18,000 prim before stalling in between
frames.

Once again,
Have a look at the picture.  It has the sim stats visible

Best Regards

Teravus Ousley
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://opensimulator.org/pipermail/opensim-dev/attachments/20071212/b66007b7/attachment-0001.html>