WALTHAM, Mass., Nov. 14 -- An M.I.T.-born start-up called
Interactive Supercomputing (ISC) today unveiled a new class of technical
computing software that, in effect, transforms powerful parallel
supercomputers into easy-to-use, interactive desktop tools for rapidly solving
large scientific and engineering problems.
ISC's Star-P(TM) is the world's first interactive parallel computing
platform. It enables scientists and engineers to code algorithms and models on
their desktops using familiar mathematical software packages such as
MathWorks' MATLAB(R), and run them instantly and interactively on SGI(R)
Altix(R) servers. Star-P eliminates the need to re-program the applications in
C, Fortran, or MPI to run on parallel computers -- a productivity-killing step
that can take months to years to complete for large, complex and
computationally intensive problems.
Star-P automatically connects desktop applications to high-performance
computers (HPCs) and parallelizes the application code on the fly, enabling
users to scale their applications across any multi-processor system or
parallel cluster in real time. Using Star-P they can tackle much larger
problems on their desktops than ever before possible, while arriving at a
solution in a fraction of the time.
"Based on recent IDC market studies, end-users won't be able to realize
the full promise of high performance computing hardware until scientists,
engineers or researchers can more easily program applications to run on
parallel architectures," said Earl Joseph, program vice president, IDC. "Star-
P is designed to address this problem by focusing on making parallel
programming an interactive approach."
For example, researchers at St. Jude Children's Research Hospital are
using Star-P to analyze MRI examinations for quantifying the impact of cancer
therapy on blood volume and flow in the developing brain of children.
Geophysicists at the University of California San Diego are using Star-P to
develop more accurate models for predicting earthquake hazards. And
researchers at the Wright-Patterson Air Force Base in conjunction with the
Ohio Supercomputer Center are using Star-P in a variety of defense-related
applications including new radar system development.
"Rather than re-coding algorithms in C or Fortran and MPI, we are looking
to Star-P to let us continue working in the high-productivity desktop MATLAB
environment, while scaling to supercomputers," said Dr. John Nehrbass,
director of computational simulations in physics at the Ohio Supercomputer
In addition to scientists, engineers and analysts doing research, Star-P
is useful to project leaders, whose primary concern is accelerating the time-
to-market of new products while lowering development costs. Star-P's ability
to eliminate re-programming in C, Fortran or MPI gives project leaders a
significant competitive advantage. Star-P also benefits infrastructure leaders
who provide HPC systems as an internal or external resource. Star-P enables
them to extend parallel computing capabilities to a wider range of users,
while increasing the return on their HPC investment.
Solving the final bottleneck to high performance computing
Star-P launches at a time when the $7 billion market for high-performance
servers for science and engineering is outpacing even the commercial server
market. Driving this demand is the fact that even the most powerful desktop
workstations can no longer handle many of the huge computational requirements
of new mathematical models and algorithms. Yet very few commercial software
applications are able to run at all on parallel computing architectures, let
Consequently, researchers who "hit the wall" with their desktop
applications are forced to stop their workflow, re-code the application in a
high-performance computing language, test the application on supercomputers in
batch mode, and then refine their work. Star-P solves this last remaining
programming bottleneck, transparently linking desktop applications directly to
parallel computers without forcing users to rewrite the code.
"In most cases, the correct approach or algorithm to a computational
problem may not be known up front. Star-P's fine-grain parallelism enables for
the first time the concept of 'computational steering' on parallel computers,
allowing users to explore and guide their coding interactively. It's a huge
productivity booster," said Pete Peterson, ISC's CEO.
Open platform for independent software vendors
Star-P is designed for high-performance computing applications in
government, defense, intelligence, academic research, life sciences, earth
sciences, energy, finance and manufacturing markets. Star-P 2.0 extends
desktop MATLAB applications to high-performance computers, with plans to
support other leading interactive desktop tools, such as Mathematica, Maple
and Python in the near future.
Star-P can also breathe new life into the thousands of commercial
science/engineering applications currently designed for desktop-only
deployment. Because Star-P is an open platform, independent software vendors
(ISVs) can easily port their existing software packages to run on parallel
architectures, eliminating the need to develop new parallel-supported
applications from scratch. It also simplifies and lowers the entry cost for
new ISV applications to enter the market for parallel computing solutions.
Star-P runs on SGI Altix servers supporting from one to 512 processors, 24
terabytes of memory, and running 64-bit Linux. Star-P will be marketed through
Silicon Graphics Inc.'s direct and indirect sales force.
"We are very excited about working with ISC to bring Star-P to the
market," said Dave Parry, senior vice president and general manager, Server
and Platform Group at SGI. "This breakthrough application, in combination with
SGI's high performance and scalable Linux systems, will give our customers a
faster path to new science, making unprecedented interactive computing power
accessible via inside their favorite desktop tools."
Pricing and availability
Star-P can be ordered today through SGI's sales channels and will ship by
year's end 2005.
Price based on ISC's U.S. list price starting at $24,000, excluding annual
maintenance and support fees. All prices subject to change without notice.
About Interactive Supercomputing
Interactive Supercomputing (ISC) launched in 2004 to commercialize Star-P,
an interactive parallel computing platform. Star-P enables automatic
parallelization and interactive execution of existing desktop simulation
applications on high-performance computers. Based in Waltham, Mass., the
privately held company markets Star-P for a range of security, intelligence,
manufacturing, energy, biomedical, financial, and scientific research