The Future of RealTime Visualization


Phil Andrews, Data Intensive Systems Mgr., Pittsburgh Supercomputing Center.

Email: andrews@psc.edu

Talk: http://www.psc.edu/~andrews/talks/ansys.html


Expect split into three distinct areas:

Split will occur because single processor speeds will be relatively invariant, distinguishing feature will be data and communication handling.

MPP Applications

Characterized by extremely large, parallel, data flow. Either externally generated, e.g., 3D data acquisition, or internally by complex codes, e.g., global weather modelling. Eaxample machines are the Cray T3E and the IBM SP2. Normally 128-1024 processors (ASCI red machine is special case, >9,000 processors).

Example 1: Realtime interpretation of MRI data.

Each MRI scan of the skull delivers 4M voxels, this must be interpreted on a one second time scale to be of use during a brain operation, in combination with a split-magnet MRI scanner.

http://splweb.bwh.harvard.edu:8000 /pages/papers/swells/mia-html/node12.html This code has run on a Thinking Machines CM2 and is now under consideration for porting to the T3E.

Example 2: Storm Prediction over Oklahoma

This large code runs on the 16-processor Cray C90 and the 512 processor Cray T3D at the PSC (see PSC home page). Data is taken each morning and processed to predict storms for that day. Visualizations have been created at the PSC and displayed on our home page as MPEG animations predicting that day's weather.

Future work:

The PSC Cray T3E will have >1TB of local, fast (fiber channel, >100MB/s) disk. We need to visualize the data on this disk in preference to post processing very large (>100GB) files. Hope to work on standardising data formats and providing a library of routines for data exploration. Sometimes rapid visualization is required before the next computational step.


SMP Applications

Characterized by moderately large data flow. Either externally generated, e.g., 2D data acquisition, or internally by somewhat complex codes, e.g., local turbulence modelling. One common configuration is an SGI Power Challenge.

Example 3: Realtime interpretation of topical voltages.

Chris Johnson's group at Utah is working on taking input from ~50 skin voltage monitors and working back from that data to heart electrical activity. This non-invasive procedure can be usedin the Dr's office. This type of capability may be appropriate for hospitals. http://http://www.cs.utah.edu/~sci

Much engineering work may take place in this regime, e.g., aircraft design. Closely linked processors and graphical devices, 4-20 processors.


Desktop Applications

Characterized by moderate data acquisition, likely to come from the internet/intranet in the future. "Data Mining" with browser-related technology will probably be a big growth market. Mostly single processor.

Typically user will extract data from a "digital library" and visualize it locally. A realtime synthesis of data mining and visualization.


The three regimes form a "pyramid" of increasing cost and complexity and decreasing numerical availability.