This month, Intel is releasing the code for its Distributed Scene Graph 3-D, which enables thousands of avatars to interact concurrently on one OpenSim region, HyperGrid Business recently reported. That would be an amazing innovation in itself, displaying many times the number of avatars now possible in Second Life or OpenSim. But even better, Intel will also put out instructions for running Scenegraph-powered regions on Amazon's cloud, which should make the technology more cost-effective (at least in many scenarios.)
The Intel tech was developed in large part by John Hurliman (known in-world as Eddy Stryker), who gave me more insights on it last year: In trials using test bots, they'd "simulate the actual load of a user by wandering around, playing animations, and downloading all of the prims and textures they see," and still managed to do that without lag. Since then, Hypergrid Business reports, the Scenegraph has been improved, enabling “thousands of users to interact in a single region.”
Sounds like spectacular news, though of course, the real test is when thousands of actual users actually try to use this. Stay tuned, and meantime, watch this Intel demo video:
Curiously, John told me last year, "There is no technical reason why you couldn't build the entire SL protocol on this architecture other than the time investment." Wonder if that's still true now. The SL protocol could sure use it.
I was at a Skye Galaxy set over the weekend (he was up after Annan Dreamscape, so I decided to stick around) and was surprised that there were as many as 90 people there at a time and I only crashed twice.
Rock Band and other rhythm games have primed the pump for acts and management (new, established and legendary) looking for alternate online outlets and venues. If we can deliver thousands of fans while maintaining rock-solid stability and performance, that's an opportunity for huge growth without changing the fundamental nature of the experience.
Posted by: Arcadia Codesmith | Monday, June 13, 2011 at 06:51 AM
Sweet... though it would be still more convincing with more diverse and fancier clothing in use rather than everyone but the POV guy dressed alike. (OTOH, when will we see a Matrix-style machinima done with this? :))
Posted by: Melissa Yeuxdoux | Monday, June 13, 2011 at 08:24 AM
The problem with many avatars in one place is two-fold:
1.) Rendering. SL avatars are very detailed, a single avatar often has more graphical data than the local scene. And the are unique, so GPU lag happens fast with just a few people. This causes VIEWERS to crash.
2.) Scripting. Avatars use many scripts, the majority of which are over their 2000k "fair share". This causes REGIONS to crash.
The avatar limit in SL is low to give people the freedom to have overly detailed avatars and ludicrous scripting. The alternative us to allow 1000's of avatars but you'd need to cap script memory and GPU usage per avatar or the clients/viewers simply could not handle it. That limits individual creativity.
In the blogged example, all the avatars are visual clones. Only one avatar needs to load, and then they are all finished loading. Similarly, I bet they aren't using 2000k script memory each, either, which already under the average usage per avatar in SL.
Given there circumstances, obviously you can handle many avatars.
Posted by: Adeon Writer | Monday, June 13, 2011 at 09:39 AM
I'm not sure who could afford the hardware to make this architecture work in practice. Or how you'd recover that cost if you did.
Posted by: Tateru Nino | Monday, June 13, 2011 at 10:05 AM
That's gonna be one hell of a party! BYOB!
Posted by: Angie Mornington | Monday, June 13, 2011 at 11:06 AM
I was guest at one of these tests about one year ago and it was a overwhelming experience, even with most of the avatars being bots and just a few (I can only guess - about 100? ) controlled by RL persons. For sure a great thing to come.
Posted by: Armin | Monday, June 13, 2011 at 01:12 PM
call up gracie kendal! it's a new twist on her 1000 avatars thingie.
Posted by: Wizard Gynoid | Monday, June 13, 2011 at 01:57 PM
Let's see it with all the avis wearing different scripts and attachments. Lab experiments prove exactly nothing. I/We want to see this for real at free open sim price levels.
I.e.; the pic shows a bunch of clones. Next?
Posted by: Ann Otoole InSL | Monday, June 13, 2011 at 02:36 PM
I was at a more recent demo. It depends somewhat on the tuning but there was negligible lag for me even with 650 avatars in a megaregion. The bots were, however, running anti-collision code and that tended to fail at high concurrency. Which was fun...
Posted by: Graham Mills | Monday, June 13, 2011 at 03:32 PM
Hmmm... if it were my project, I'd handle avatar customization by implementing a strict avatar rendering routine where priority is based on proximity, with a method of designating the on-stage act as the primary override.
It means 975 of the 1000 attendees may never see your supercool rainbow glowing body armor... but they're not there to see you, they're there to see the show. If you want everybody to see your fabulousness, book your own performance.
Posted by: Arcadia Codesmith | Tuesday, June 14, 2011 at 07:55 AM
I doubt there'd be many times you'd need to worry about 1000 avatars, but if this helps us get 50 or 100 decked out avs in the same sim with no lag it will be a big help.
Posted by: David Cartier | Thursday, June 16, 2011 at 08:44 PM