The Access Grid at Vislab

1
The Access Grid at Vislab

• Chris Willing
• chris_at_vislab.usyd.edu.au
• Vislab
• University of Sydney, Australia.

2
Today

• Background
• Differences to traditional vc
• Vislab Implementation ATP
• Vislab Implementation Physics
• Virtual Venues
• Differences to ANL implementation
• Future Work

3
Background

• VISLAB visualisation, high perf. Computing
• Labs at Physics, ATP
• Bernard Pailthorpe at SDSC in 2000
• High resolution displays
• AG node
• Start April 2001 for SCGlobal, November 2001

4

Hyogo Prefecture (Japan) Police Command Center
- reference site
Tiled rear projection array of 6 Model 200 ILA
Projectors
5
SmartSpaces, Stanford
6
PowerWall, Princeton
8x commodity projectors
7
High Density Tiled Display, SDSC
http//vis.sdsc.edu/research/tileddisplay.html
3x3 array with common light source
8
(Background)

• VISLAB visualisation, high perf. Computing
• Labs at Physics, ATP
• Bernard Pailthorpe at SDSC in 2000
• High resolution displays
• AG node
• Start April 2001 for SCGlobal, November 2001

9
Display, VisLab 2001
3x1 array, 3840x1024 pixels, single light source
10
Application in archaeology Angkor
11
(Today)

• Background
• Differences to traditional vc
• Vislab Implementation ATP
• Vislab Implementation Physics
• Virtual Venues
• Differences to ANL implementation
• Future Work

12
Differences 1. traditional

• Each node sends image stream(s) to MCU which
  decides what is output to each node
• TV style

• MCU

13
Differences 2. access grid

• Each node sends image streams to a multicast
  group address
• Each node sees all other sources

14
but if Each node sees all other
sources then screen overload
e.g. 10 other sites, each with 3 cameras gt 30
video streams
15
Normal screen
16
Widescreen 5120x1024
Normal screen 1280x1024
17
Summary of Differences

• limited vs. rich user experience
• complex (studio) vs. simple (pc)
• proprietary vs. open
• expensive vs. cheap
• but richness requires pixels

18
Technology diversion 1

• Unicast
• 3 streams to 7 sites 21 video streams

19
Efficient but who pays?

• Multicast
• 3 streams to 7 sites 3 streams (mostly)

20
Today

• Background
• Differences to traditional vc
• Vislab Implementation ATP
• Vislab Implementation Physics
• Virtual Venues
• Differences to ANL implementation
• Future Work

21
Vislab Implementation (ATP)

display
projectors (monitor)
video
cameras
audio
mics
Gentner
PA
22
Access Grid Sydney
Preparing for SC-Global test cruises
A/G-Sydney
23
ATP display
Projected display 3840x1024
Mon.
display
network
24
ATP video capture
screen
camera
table
video
network
projectors
25
ATP audio capture
26
Gentner echo control
echo delay due to - distance
- application
buffering
27
Vislab Implementation (Physics)
28
Virtual Venues

• ANL runs Virtual Venues Server
• APAG server at http//venues.ap-accessgrid.org
• Virtual rooms characterised by facilities
• Video multicast group address
• Audio multicast group address
• MUD location (back channel link for participants)

29
Differences to ANL

• Linux only
• stability
• sync with vvd (not CORBA event handler)
• potential for compute clustering (openMosix)
• Graphics cards (nVidia Quadro4)
• use well known visualisation apps
• e.g. performer, openInventor, openDX
• potential for video clustering (chromium)

30
Video clustered AG

31
Performer with AG
32
High Resolution with AG
33
Future

• Video clustering, stereo viewing
• Shared event stream
• Higher resolution (PAL, HDTV)
• H263, hardware MJPEG
• External machine capture
• Coexistence with H323 (via VRVS)

34
Links

• http//www.vislab.usyd.edu.au/research/accessgrid/
  
• http//www.vislab.usyd.edu.au/research/display/
• http//www.accessgrid.org
• http//www.scglobal.org
• http//vis.sdsc.edu/research/tileddisplay.html