BLOOMINGTON, Ind. -- High performance computing (HPC) has brought about many revelations, from the likely origins of the moon (a collision between a celestial body and Earth) to the constant, accelerating expansion of the universe. Before HPC (also known as supercomputing), it was next to impossible to crunch the numbers behind these revelations within a lifetime.
The National Science Foundation (NSF) recently awarded Indiana University's Center for Research in Extreme Scale Technologies (CREST) two grants totaling nearly $430,000. Together, the grants will help develop a skilled HPC workforce and improve technologies that could make robots more useful in a number of areas, including manufacturing and surgery.
The first NSF grant totals nearly $200,000 and provides funding to develop an online HPC course to improve America's reach in the little-tapped world of exascale computing. Exascale computing is the gold standard of all HPC, with a capacity of at least one exaflops, which is 1,000 times more powerful than the first petascale supercomputer from just five years ago. Few people are qualified to work with computers of this scale. CREST Executive Associate Director and Chief Scientist Thomas Sterling and his partner, CREST Director Andrew Lumsdaine, aim to change that with this course.
"There is a dearth of experts in computational sciences and in HPC systems design in this country, and we have to reverse that, or frankly it will have a negative economic impact on the nation," Sterling said.
An online HPC course has its advantages, Sterling said. The course does not meet at all, so schedule conflicts are not an issue. Lectures are recorded and course work is available at any time, so students can complete the curriculum at their convenience. Because the lessons are recorded and sourced from anywhere, the online HPC course will draw from some of the world's finest HPC experts to teach the class. CREST also hopes to make the class available to students at institutions that are not equipped to provide such courses.
"Unfortunately, the quality of your educational life is often predetermined by your demographics and socioeconomics, and that's wrong," Sterling said. "Everyone should have equal access to quality education so they have maximum choice, and that's one of the goals here."
A second NSF grant totaling $230,000 will help IU further develop technologies that could ultimately make robots more useful. IU has designed a multiprocessor execution model known as ParalleX, which tells supercomputers how to carry out computations. The grant will help researchers make ParalleX suitable for exascale computing, enabling the largest supercomputers in the world to run on multiple processors. To do that, researchers must first make several improvements to ParalleX, namely to its reliability, energy use, debugging capability, and ability to manage how a computation is distributed across an exascale computing platform.
Researchers must also make it so ParalleX can understand the notion of time. Combined with the other advancements, this understanding could move ParalleX into the realm of embedded computing, which can theoretically enhance a robot's ability to perform certain tasks. For example, technology like the Mars rover, Curiosity, could one day benefit from ParalleX. Curiosity is the size of a small car and run by only one processor. Each real-time program is written one instruction at a time. While this is a very reliable method, it is also labor intensive. It would be much less so if multiple processors could be used, which is something ParalleX – once modified for exascale computing and real-time control – could help with, according to Sterling.
"Our ability to include real-time semantics means that we can offer an alternative approach to real-time control that takes advantage of much more processing power and performance," Sterling said. "If you combine that with our dynamic approach to reliability, it also means that it can provide greater confidence and reliability to the process that's taking place. I really think it allows us to complete the software abstraction we need in order to make robots more widely useful."