[sldev] Scalable Sim Question
lenglish5 at cox.net
Sat Aug 9 13:02:57 PDT 2008
Sheet Spotter wrote:
> I would be curious how extensively the existing SL server exploits the
> use of multi-threading and multi-processor.
> Is there any opportunity to distribute the functional blocks across
> multiple threads or processors?
> For example, could the physics be run on one processor, while the
> communications (chat, IM, notices, etc.) runs on a separate processor,
> and the scripts ran on yet another processor?
> I realize there are dependencies and overlap. Chat distance relies on
> the location of each avatar, which is calculated by the physics
> engine. Similarly, scripts frequently rely on position information for
> performing sensor scans, etc.
> Are physics, communications, and scripts even the three most important
> functional blocks? Hmmm…Texture downloads probably figure prominently
> in the equation.
The first part at refactoring how the grid works is being tested right
now with the Open Grid Public Beta testing.
Logging into the test metaverse and TPing between grids is now being
accomplished by the client interacting with the "Agent Domain," a server
which eventually will be analogous to an ISP for your avatar account(s)
and it will handle login, group IM, persistent inventory, etc., while
the local simulator will handle physics, local chat and geometry issues.
The eventual division will be "Agent Domain" vs "Region Domain" where
each is responsible for a certain subset of services that the SL client
now receives exclusively through the simulator.
Within the AD and RD there could easily be multiple servers responsible
for various servics, but that should be transparent to the client. There
may be other "Domains" defined as time goes on to logically distribute
the responsibilities a bit more.
More information about the SLDev