[Opensim-dev] I'm new here, and I want to improve scalabilty.

[Sean Dague]

On Fri, Nov 16, 2007 at 11:07:57AM +0900, Johan Berntsson wrote:
> Hi,
> 
> I have just started to look at OpenSim, and I'm especially interested in
> scalability issues. In my preliminary tests the server is able to handle
> up to 20-25 clients in a simple world using ODE physics, and I would
> like to find ways of increasing this number to 100 clients or more. The
> main limitation at the moment seems to be the CPU load of physics
> calculations, but even with the basic physics there are limitations that
> don't appear to be CPU related.

If you look back on the list a week or so you'll see some early results
that both Mic and I listed on our scalability findings so far.  We
seemed to hit different hot spots at the moment.

Real physics is always going to be high on the list of hot spots.  "It's
hard and stuff."  One of the things that was previously talked about was
creating an OpenSim stress harness that would create lots of libsl bots
that would run around randomly.  That + profiler would help force out
more hot spots.

That being said, OpenSim code is still a long ways from feature
complete.  There are loads of packets not implemented yet, as well as
minimal scripting.  Be careful of premature optimizations at this point.

> Before I start modifying the code I would like to know what has been
> done and tried before, and I'm hoping that we can start a discussion on
> what the present scalability bottle-necks are and ways to improve the
> code.I have worked with parallel and distributed computing in other
> applications, and I'm familiar with C#. My nickname on IRC is joha1.

I'm personally starting to look of the UDP in / out path in ClientView
and clean it up a bit, hopefully shaking out some perf gains in the
process by being smarter about what packets are sent / resent.  No idea
if that will have noticable overall effect (it should a little bit at
least), but the motivation is more from general cleanup.

Sticking profilers on the current stack (much easier to do in mono than
in .net) would be very helpful to start getting real data on where
hotspots are.  Please share those results on list, as I'm sure we'd all
find it useful.

     -Sean

-- 
__________________________________________________________________

Sean Dague                                       Mid-Hudson Valley
sean at dague dot net                            Linux Users Group
http://dague.net                                 http://mhvlug.org

There is no silver bullet.  Plus, werewolves make better neighbors
than zombies, and they tend to keep the vampire population down.
__________________________________________________________________
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://opensimulator.org/pipermail/opensim-dev/attachments/20071116/ffaa2ff5/attachment-0001.pgp>