Dataset Massive Visualization
Aiichiro Nakano
Collaboratory for Advanced Computing & Simulations Dept. of Computer Science, Dept. of Physics & Astronomy, Dept. of Chemical Engineering & Materials Science University of Southern California Email: [e-mail]
Immersive & Interactive Visualization
Billion-atom walkthrough
Parallel & distributed Atomsviewer
Reduced data
W A N D
Graphics server
PC cluster ImmersaDesk
User position
Locality in Data Compression
Massive data transfer via wide area network: 75GB/step of data for 1.5 billion-atom MD! Compressed software pipeline Scalable encoding: Store relative positions on spacelling curve: O(NlogN) O(N) Result: Data size, 50Bytes/atom 6 Bytes/atom
14 4 3 2 1 7 8 9 5 6 10 13 12 11
Data Compression for Scalable I/O
Challenge: Massive data transfer via OC-3 (155 Mbps) 75 GB/frame of data for a 1.5-billion-atom MD! Scalable encoding: Spacelling curve based on octree index
x y z R = 1 1 0 =0 0 0 =1 0 0 = 101 001 000
3D list map preserves spatial proximity
Spacelling-Curve Data Compression
Algorithm:
1. 2. Sort particles along the spacelling curve Store relative positions: O(NlogN) O(N) Adaptive variable-length encoding to handle outliers User-controlled error bound
Result:
An order-of-magnitude reduction of I/O size: 50 6 Bytes/atom
Data Locality in Visualization
Octree-based fast viewfrustum culling Probabilistic occlusion culling Parallel/distributed processing
Reduced data
W A N D
Graphics server
PC cluster ImmersaDesk
User position
Interactive visualization of a billion-atom dataset in immersive environment
Hierarchical Abstraction
l=0 l=1 l=2 l=3
2D example Use of an octree
Larger clusters for longer distances Recursively subdivide the 3D space to form an octree
Visibility Culling
View frustum culling Higher Depth
Viewpoint
Occlusion culling
Octree-based View-Frustum Culling
Use the octree data structure to efciently select only visible atoms Complexity Insertion into octree: O(N) Data extraction: O(logN)
Probabilstic Occlusion Culling
Remove atoms that are occluded by other atoms closer to the viewer Regions farther away from the viewer is more likely to be occluded than one in front of the viewer
Draw fewer atoms per region as the distance of a region from the viewer increases: value visibility v(x) for region x Recurrence along the view line # 1 x=0 vx = $ % f (Dx " , v x " ) else Dx = density of region Run time adaptation
v " = f (user speed) # v x x
Results of Probabilstic Occlusion Culling
Original Probabilistic Difference
68% fewer objects 3 frame rate
Multiresolution Culling & Rendering
Per-octree node operations: Frustum culling Probabilistic occlusion culling Per-atom operations Multiple levels-of-detail Occlusion culling (per-object, per-octree node)
Without multiresolution With multiresolution
.94fps - 90,000 particles
3.2fps - 4,500 particles
Outow pathways of optic nerves from the retina of a rabbit eye (Experimental data by C. Burgoyne & R. Beuerman, LSU Eye Center)
Distributed Architecture
RENDERING & VISUALIZATION MODULE
RENDERING SYSTEM PER-ATOM OCCLUDER
USER POSITION
NEAR COMPLETE LIST OF VIEWABLE ATOMS
OCTREE BASED DATA EXTRACTION MODULE
PROBABALISTIC OCCLUSION CULLING MODULE
REGIONS OF INTEREST
TCP/IP SOCKET
Parallel Octree Extraction
PC Cluster Nodes
Bounding Shells of Equal Volume
Individual copies of the octree with each node Spherical extraction by the use of shells of equal volume Load balancing due to the equal use of each processor for extraction
Latency Hiding
Individual modules are multithreaded to reduce network or module latency Minimize latency due to inter-...