[Opensim-dev] New MOSES Physics Video (UNCLASSIFIED)

Wed Apr 15 14:20:55 UTC 2015

Classification: UNCLASSIFIED
Caveats: NONE

1.  This is an excellent point that illustrates just how wickedly complex our problem is.  The short answer is "yes", we could do it locally.  However, the distribution of the shared experience is a hard requirement and the final state of the bricks must be accurately portrayed to all client connections.  One way to handle this is to treat this like a state machine and make sure all clients have precisely the same start conditions and then apply the same physics calculations.  Theoretically, we would have the same end conditions.  Afterward, we could queue the state changes of all the object and then have an arbitration process at the server make sure everything is actually at the same global location and orientation.

This is a complex issue as it is contrary to the current world state model and would require the server to temporarily relinquish control of prims during the physics action, then take it back after steady state.

3.  Our PhysX work is simply to ensure the physics calculations we need in the 83ms interval is actually completed and not dropped.  It is address any "rubber banding" effects you see.  It should also address jitter.  It is a balance between update rate, network saturation, and QoS.

4.  The DSG middleware approach gave us many insights into how a truly distributed simulator could work.  That was extremely valuable work for us and we were privileged to have excellent collaborators in Intel and a strong community of MOSES users willing to help us test. 

v/r -douglas

Douglas Maxwell, MSME
Science and Technology Manager
Virtual World Strategic Applications
U.S. Army Research Lab
Simulation & Training Technology Center (STTC)
(c) (407) 242-0209

-----Original Message-----
From: opensim-dev-bounces at opensimulator.org [mailto:opensim-dev-bounces at opensimulator.org] On Behalf Of Mister Blue
Sent: Wednesday, April 15, 2015 12:21 AM
To: opensim-dev
Subject: Re: [Opensim-dev] New MOSES Physics Video

1. Is the explosion of blocks or prims something that the viewer can handle? Or is this too tricky to make happen with the wide variations of machine running viewers and even more so that soon the viewer will run in a web browser? 
It is something a viewer could do but I don't know what mechanism could be used in the existing viewer.

2. Is there a way to make the explosion an overlay streaming event that runs over the current screen? - Just a crazy idea.... I am thinking of this more on a browser-viewer as that needs to run on devices that would have issues processing all that...
In interesting idea but, again, would be viewer modifications.

3. Is it possible to make OS Physics run faster than 11FPS?
Yes. The default configuration is for the simulator heartbeat time to be 11FPS and for the physics engine to be in sync. You can change the simulator heartbeat time but there are some known problems with that (too many assumptions in some places). If you run BulletSim on its own thread, you can speed up BulletSim but that would just make more updates.

4. It seems that the number of avatars exponentially changes the workload here. Maybe a graphics server could be designed as a sub service to handle that type of load, maybe running on a GPU instead of a CPU? It just seems to me that with all the other things that the region server has to do, offloading some of the heavy lifting would be a good thing. Maybe it is time to think of an OS "Pro" level of setup that separates the workload a bit more would be a good idea

That was the design of DSG. For client connections, the simulator fed on client manager who multiplied the connection to multiple viewers. So you could have a simulator feeding 10 client managers who were each feeding 30 viewers for a total of 300 connections. DSG also run script and physics servers to off load the central server from the computation from those operations.

== mb

On Tue, Apr 14, 2015 at 1:49 PM, Dahlia Trimble <dahliatrimble at gmail.com> wrote:

	I believe there is a means in the LLUDP protocol to stuff many updates for many objects into a single packet, though I'm not sure OpenSimulator is smart enough to do it in your simulation. It may be a way to improve networking performance quite a bit when may physical objects change velocity during the same simulation frame.

	On Tue, Apr 14, 2015 at 1:44 PM, steve l <salbiedermann at gmail.com> wrote:

		Hi!

		Robert- Thanks for the answer and the thought you put into it!

		So I am going to play dummy (Not far from the truth!) here. This means that we need to do re-writing on several parts of OS to speed things up and eliminate bottlenecks. a couple of questions then.

		1. Is the explosion of blocks or prims something that the viewer can handle? Or is this too tricky to make happen with the wide variations of machine running viewers and even more so that soon the viewer will run in a web browser? 

		2. Is there a way to make the explosion an overlay streaming event that runs over the current screen? - Just a crazy idea.... I am thinking of this more on a browser-viewer as that needs to run on devices that would have issues processing all that...

		3. Is it possible to make OS Physics run faster than 11FPS?

		4. It seems that the number of avatars exponentially changes the workload here. Maybe a graphics server could be designed as a sub service to handle that type of load, maybe running on a GPU instead of a CPU? It just seems to me that with all the other things that the region server has to do, offloading some of the heavy lifting would be a good thing. Maybe it is time to think of an OS "Pro" level of setup that separates the workload a bit more would be a good idea.

		These things always get me thinking...!

		Steve LaVigne

		A Dimension Beyond, Inc.

		www.adimensionbeyond.com

		On Tue, Apr 14, 2015 at 10:45 AM, Adams, Robert <robert.adams at intel.com> wrote:

			I don’t think the only problem is finding a physics engine that can handle 240 moving objects. Another is optimizing the updates from the physics engine.

			Think of the whole pipeline: the physics engine computes interactions and new locations/rotations for each object. That position update is sent to the simulator. The simulator updates the object data structures and sets an update flag. The location/position update is noticed and an update packet[1] is created and placed in output queues for each viewer. At some time, the packet is transmitted to each viewer.

			The update processing time can easily be more than the physics engine time.

			The OpenSimulator physics engines are run 11 times a second so they generate 11 position updates a second for each moving object. So, even an efficient physics engine will generate (240 * 11) updates per second which then turn into (240 * 11 * numberOfAvatars) packets sent per second.

			There are many optimizations possible in this chain.

			-- mb

			[1] This is technically wrong for the current version of OpenSimulator. For the technically inclined, an ‘update needed’ packet is put in the output queue and the actual packet to transmit is created when it is time to send the update. This is done because the update output packet queue can get long and the position/location information can be stale if multiple updates are in the queue. Only one ‘update needed’ packet is put in the queue and the current object location/rotation is put in the transmitted packet at the time of transmission.

			From: owner-moses-list at lists.mitre.org [mailto:owner-moses-list at lists.mitre.org] On Behalf Of steve l
			Sent: Tuesday, April 14, 2015 8:09 AM
			To: Michael Emory Cerquoni
			Cc: opensim-dev at opensimulator.org; moses-list at lists.mitre.org
			Subject: Re: [Opensim-dev] New MOSES Physics Video

			Hi!

			An excellent video on the physics of exploding grenades and the wall.  On the OS Dev list Mister Blue has an excellent observation that the server crashes are due to the extreme amount of changes that have to be sent to every avatar. His idea of a client side solution might just be a good one. In the end is there any way that OpenSim can handle more events than that in it's present form? Is there any physics engine that can handle 240 moving scripted objects moving at once without lag?

			If we could get OS to the point that it would handle this load easily, we would have all our load issues solved!

			Steve LaVigne

			On Tue, Apr 14, 2015 at 6:33 AM, Michael Emory Cerquoni <nebadon2025 at gmail.com> wrote:

			Could these test scripts be shared so testing against other engines can occur as well, I would be interested to see how this same test goes against ODE and BulletSim as well.

			On Tue, Apr 14, 2015 at 8:18 AM, Maxwell, Douglas CIV USARMY ARL (US) <douglas.maxwell3.civ at mail.mil> wrote:

				Good Morning, as you all know the MOSES developers are working on PhysX integration into the Open Simulator to support functionality currently not possible in the platform.  We are a methodical group and a couple months ago I asked one of our interns to work with the developers to create a series of baseline physics behavior case studies.  The first case study is a destructible wall caused by an explosive charge.  This wall is composed of blocks that are tested at a high density and a low density to simulate different destruction effects.  

				The goal here is to eventually have all of the prims in the sim loaded with the scripts needed to react to any type of random explosive charge set by the participants in the training scenario.

				The video can be found below:

				https://youtu.be/jSofWcwWi7g <blockedhttps://youtu.be/jSofWcwWi7g> 

				Your feedback is welcome.

				Observations:

				1)  Current limitations of the open simulator prevent us from expanding the tests beyond a simple wall.

				2)  The scripts exercise the engine well and expose limitations between the sim frame rate and the physics frame rate.

				3)  It is easy to crash the sim with this demonstration, especially if more than a handful of people are present (more than 3-4 client connections).

				Douglas Maxwell, MSME
				Science and Technology Manager
				Virtual World Strategic Applications
				U.S. Army Research Lab

				Human Research & Engineering Directorate
				Simulation & Training Technology Center
				(c) (407) 242-0209 <blockedtel:%28407%29%20242-0209> 

				_______________________________________________
				Opensim-dev mailing list
				Opensim-dev at opensimulator.org
				http://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev <blockedhttp://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev> 

			-- 

			Michael Emory Cerquoni

		_______________________________________________
		Opensim-dev mailing list
		Opensim-dev at opensimulator.org
		http://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev <blockedhttp://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev> 

	_______________________________________________
	Opensim-dev mailing list
	Opensim-dev at opensimulator.org
	http://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev <blockedhttp://opensimulator.org/cgi-bin/mailman/listinfo/opensim-dev> 

Classification: UNCLASSIFIED
Caveats: NONE

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 5629 bytes
Desc: not available
URL: <http://opensimulator.org/pipermail/opensim-dev/attachments/20150415/878f7ce1/attachment.bin>