Article courtesy Hewlett-Packard
Like scientists and engineers in many fields, climatologists crave interactive tools with which to watch their field data and simulations
evolving in real time. Watching natural processes evolve over
time -- whether on a scale of microseconds or millennia -- can spark pivotal
insights, giving rise to an intuitive grasp of these processes that
can't be readily obtained otherwise.
Steer your visualization
Such a tool has recently been developed by John Clyne, a software engineer in the Scientific Computing Division (SCD) at NCAR. Clyne has used a Hewlett-Packard Exemplar
server to transform batch-mode visualization software into a real-time
"Graphics workstations can't do time-evolving volumetric imaging for us
because hardware-accelerated systems render images by calculating
polygons -- essentially triangles," says Clyne. "Typically our data
represent turbulence in the atmosphere, oceans, or sun, and volumetric
data of this type don't support polygonal rendering."
For this reason, he says, "No commercial product renders images of
time-evolving volumetric data like ours, certainly not in real time."
It all has to be done in software, such as VOLume-rendering SHell
(Volsh), which relies on the computational muscle of the computer's
High performance, easy development
NCAR's vector-based supercomputers, even when running in parallel mode,
don't perform such calculations interactively in real time. Most of the
work that's been done in parallel volume-rendering on these computers
has focused on rendering single time-steps -- a single data set
essentially producing a snapshot of a dynamic process.
many of these in batch mode, scientists can produce an animation of
their data. But, Clyne says, "In batch mode, you run it out and hope
you've got the right thing. You can look only at the result. If it's
not what you want, you have to go back and do it again."
By running the new version of Volsh on the Exemplar server, NCAR
scientists will be able to interact with the animation as its taking
place. Clyne describes the experience: "You can sit there and steer
your visualization. You can move forward or backward in time, zoom in on a feature, or change viewpoints -- observe from a different angle -- all in real time."
Despite Volsh's interactive, intense computational demands, he says,
"The Exemplar server is doing everything we need it to do." The system
architecture's low latencies and low overhead contribute to the high
performance required to run Volsh interactively.
Coming next: GUI
"Even when we use
processors in multiple nodes, we see a near-linear performance
increase. And we haven't done anything to our code to optimize memory
layout. We just pump data into memory and let the machine handle it."
Because these data evolve through time, says Clyne, "it throws a whole
new wrinkle into the problem. Now you not only need intensive
computation but also intensive I/O. You've got to be able to feed the
The Exemplar server provides the input/output bandwidth to
handle NCAR scientists' large data sets in real time. And I/O is
scalable along with Exemplar processing power, memory, and other
aspects of the system.
Further, Clyne says, "The shared-memory programming model makes the
Exemplar server very easy to work with." This shortens development
time so NCAR scientists can begin interacting with their visualizations
much sooner than would otherwise be possible.
Clyne is developing a graphical user interface with which users will
interact with the new Volsh version. When he completes this last
development step, interactive Volsh will be placed in the public domain.
Like earlier versions of Volsh, it will be published on the Web at
www.scd.ucar.edu/vg/Software/volsh for use by other climatologists as
well as by scientists and engineers in other disciplines.
Volsh could be used in any science that must deal with enormous
volumetric data sets," Clyne says.