| News |
| High Productivity Computing Systems |
| DARPA HPCS Program |
| Dr. Frederick Johnson |
| The key to understanding the DARPA (Defense Advanced Research Projects Agency) HPCS (High Productivity Computing Systems) program is that the ‘P' in HPCS stands for productivity and not petascale or performance. While execution performance is an essential component of system productivity, the HPCS program also targets major improvements in ease-of-use and development time activities to enhance productivity for users of HPCS systems. |
In the latter part of the 1990's almost all high-performance computing (HPC) systems were based on clusters of commodity nodes, and the programming model for these systems was MPI coupled with a standard computing language such as C, C++, or Fortran. Many mission-critical applications of the federal government were difficult to implement on these machines, and after many detailed studies, DARPA was charged to develop the HPCS program. The program began in 2002 with an overall goal of providing a new generation of economically viable, high productivity computing systems for the national security, mission agency, and industrial user communities by 2010. Specifically, HPCS systems are expected to impact HPC in many areas, including:
- Performance(time-to-solution): speed up critical applications byfactors of 10 to 40;
- Programmability (idea-to-firstsolution): reduce cost and time for developing application solutions;
- Portability: insulate application software from system specifics;
- Robustness: protect applications from hardware faults and system software errors.
|
| While we have many metrics for evaluating execution performance, there are few metrics for development time productivity. However, an expert in the intelligence community has said that HPCS systems would be successful in his area if they increased the number of people able to effectively use the systems by a factor of ten. This is one practical measure of development productivity. |
| The HPCS program has three phases. Phase one was a one-year concept study phase, in which five supercomputer vendors Cray, HP, IBM, SGI, and SUN participated.
Phase two is a three-year research and development period, in which Cray, IBM, and SUN participate. Phase two will conclude this summer, and phase three is about to commence. All three phase two vendors have submitted proposals, which are now under evaluation. |
| Currently, few details concerning the HPCS architectural designs are available publicly, but more and more details will emerge as phase three progresses. This information will allow application developers and system software experts to prepare for these systems. |
The system vendor activities represent the principal focus of the HPCS program. However, the system has several other components, as well:
- Productivity Assessment – The productivity assessment team conducts research in development and execution time productivity metrics and benchmarks. Part of these activities is the HPC Challenge benchmark suite and result tables, which augment the Top500 Linpack results with several metrics that are more meaningful to the HPC community
- High Performance Language Systems – Each of the phase two vendors is developing and studying new languages for simplifying the programming of petascale parallel systems. Cray's language is Chapel, the IBM language is X10 and SUN's language is Fortress. Additional information about these languages is readily available on the Internet.
- Application Development – Key mission partners will undertake petascale application development activities to accelerate the availability of petascale applications for HPCS systems. It is likely that certain SciDAC applications will have the opportunity to participate in this process as these ongoing plans mature.
|
| The DOE SC and NNSA have been active in supporting the HPCS program from its inception. Representatives from these two organizations have participated in proposal evaluations for all three phases, and in all of the quarterly progress reviews with each vendor – over 50 reviews. Much more importantly, SC and NNSA are HPCS mission-partners, having provided financial backing, with plans to continue funding. Mission-partner application needs have been incorporated into the requirement specifications for the phase three systems. SC has funded research activities in development and execution time productivity in phase two, and that will continue and possibly expand this effort in phase three. |
| With strong participation from a wide spectrum of government agencies, including SC, NNSA, NSA, NASA, NSF, NRO and the DOD Modernization Program, the HPCS program exhibits innovation in interagency coordination, as well as HPC hardware and software. It is very likely that the HPCS systems that emerge in 2010 will significantly impact agency mission-critical applications. |
| Dr. Frederick C. Johnson, Senior Technical Manager for Computer Science, Office of Advanced Scientific Computing Research. |
Further Reading
http://www.darpa.mil/ipto/programs/hpcs/index.htm
http://www.highproductivity.org/
http://icl.cs.utk.edu/hpcc/
|
| |
|
