Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Myriad: Scalable VR via Peer-to-Peer Connectivity, PC Clustering, and Transient Inconsistency Benjamin Schaeffer Peter Brinkmann George Francis Camille Goudeseune Jim Crowell Hank Kaczmarski Integrated Systems Lab University of Illinois Urbana-Champaign

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Syzygy: PC Cluster Virtual Reality In production use at 3 PC Cluster CAVEs at the Integrated Systems Lab. 3.5 years of continuous psychology experiments using PC Cluster CAVEs. Over 5 years of development. First release September ,000 words of documentation. Inexpensive High performance Focus on quality cluster synchronization Cross-platform: Windows, Linux, Mac OS X, SGI OPEN SOURCE

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Peer-to-Peer Reality (Terms) Myriad operates using a network of reality peers, each of which holds a scene graph. Scene graph updates are messages that travel the peer network, altering peers as they go. The network uses connection feedback to reduce message traffic between peers. Further message filtering (on peer connections) occurs because of fine-grained sharing and reality maps, which enable complex relationships between peers, such as subjective views. Fine-grained sharing means that the sharing properties are controllable down to the level of the individual scene graph node. A reality map relates scene graph nodes in connected peers. Transient inconsistency means that connected peers can drift into and out of agreement with one another, based both on user and system action. The system is highly threaded and so must be fully thread safe.

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Applications Collaborative editing of 3D worlds, particularly of large scale environments. Potential uses in the animation industry, for instance. 3D collaboration across systems with very heterogeneous network, compute, and graphics resources. The future… what would terascale or petascale virtual reality look like? How about a decentralized, ad hoc network with sufficient power and storage to recreate the world?

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA WorldWideCrowd: Demo Configuration Can be easily set up because the peer network is scriptable in Python.

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Multiple views 3D manipulation Python-based manipulation Many-to-Many Collaboration

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Peer 1 Peer 5 Peer 2 Peer 4 Peer 3 Peer connection Messages discarded here. Peer Network: Handling Connection Cycles

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Peer 1 Peer 2 Consistency Process Time Transient Inconsistency

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Late-Joiner State Transfer: Transient Inconsistency Avatar can move while state is transferred in background. Thread safety Time

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Myriad Scene Graph points1 normals1 colors1colors2 drawable (triangles) drawable (triangles) points1 normals1 colors1 points1 normals1 colors2 Node Types transform texture bounding sphere billboard visibility blend light material graphics state normal3 color4 tex2 index points drawable persp camera viewer Geometry

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Node levels 1. Structure 2. Stable 3. Optional 4. Transient Data flow between top nodes Stable Outbound filter: Structure Outbound filter: Transient Reality maps not necessarily “onto”. Reality Map: Node Levels and Outbound Filters Fine-grained Sharing

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Peer 1Peer 2 Map filter: Transient Limb transform nodes have transient level. Message Filtering: Transient Nodes Connection feedback and Fine-grained sharing 1. Connected peers send one another their framerates. 2. Each transient node keeps track of when it last sent an update to each connected peer. 3. Message send rate throttled to framerate by filtering.

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA A merging algorithm is needed because graph isomorphism is a difficult problem. The system must support partial sharing (subjective views, transient inconsistency, connection feedback). 1. Assume parent node is mapped. 2. Given a local child node. a. If it is not a name node, search below the parent node’s image for an unmapped node of the same name and type. b. If it is a name node, search below the parent node’s image for an unmapped name node. 3. If a node is found on the remote peer, extend the map. 4. If no node is found, create a new node on the remote peer and extend the map. The mapping process can also transfer node contents (depending on node level and mapping level). Local Peer Remote Peer (image) Mapping direction Constructing Reality Maps: Merge Algorithm

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA 1. When a peer receives a message, update the message’s internal history. Discard the message if it has already visited this peer. 2. If the message is from a mapped node, use the reality mapping to determine its local destination, altering the message (which stores its destination node ID). 3. Discard if the message goes to a locked node, but its origin is not the lock’s owner. 4. Update the local peer’s scene graph. 5. If the message creates or deletes nodes, update the inbound reality map (for the incoming connection) and the outbound filters (on all other connections). Note that the “merge” algorithm uses node creation messages. 6. Return node mapping information back to the sending peer upon node creation. 7. Send to connected peers if conditions on node level/ mapping level and transient node updates are met. Inbound Reality Map (translates between node IDs) PeerConnection Message Filtering Algorithm Outbound filter (discard based on node level)

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Subjective Views Different avatars Transient Inconsistency

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Subjective Views Different Animations Transient Inconsistency

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Root Material node Transform nodes Sphere geometry Reality Map: Inserting Material Nodes

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Root Material node Sphere geometry Transform nodes New material nodes Peer connection Peer 1 Peer 2 Reality Map: Inserting Material Nodes

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Peer 1Peer 2 Reality Map: Inserting Transform Nodes Torii rotate in each peer.

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Peer Connection Scalability: Peers on Underpowered Computers Underpowered computer

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Time Peer 1 Peer 2 Scalability: Peers on Underpowered Networks Share only one avatar at a time, varying the avatar over time. Slow network

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Virtual Reality View CAVE screen junction… picture taken away from viewpoint.

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Smooth Pan Down into the Crowd Connection feedback, scalability Time

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Simulation peer Motion culled from display peer Display peer Display wall: Connection feedback (motion culling)

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Drawn for Tile 1Culled from Tile 1 VR View: Scalable Tiled Wall

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Editing this avatar. Interactive Editing at the Python Prompt Partial database listing.

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA New texture.Command to make change.

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Different transform.

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Entirely new avatar.

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Changing individual point position. Fine-Grained Sharing

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Python Collaboration Thread safety Peer 1Peer 2

Myriad Scalable Scene Graph, VRST 2005, Monterey, CA Getting the Software Source code and data to recreate the experiments described here can be downloaded from DVDs with precompiled SDKs, source code, many demos, etc. are available here at VRST. Please ask! Everything necessary to recreate our lab’s operation. Full documentation available on the lab website.