Reaching Out to the Next Generation of HPC Users
SciDAC Outreach Center provides vital conduit
David Skinner, Ph.D.
Everybody, it seems, is talking about the importance of high performance computing to the nation’s scientific, economic and educational competitiveness. From the
 |
president to Congress, government agencies to the Council on Competi-tiveness, the importance of helping new research communities gain access to HPC resources is widely discussed. But without connections to the supercomputing community, it can be a challenge just knowing where to start.
To provide the initial stepping stone, the U.S. Department of Energy’s (DOE) Scientific Discovery through Advanced Computing program (SciDAC, www.scidac.gov) last year launched the SciDAC Outreach Center. This new Center will provide more structured outreach efforts, building on less formal outreach by other SciDAC Centers. For example, the Applied Partial Differential Equations Center (APDEC) has made its software available to researchers at other institutions, resulting in the simulations accompanying this article. Tools developed by APDEC at Lawrence Berkeley National Laboratory (LBNL) are being used to advance research in areas such as fusion energy, combustion, astrophysics and oceanography.
Although the Outreach Center’s primary focus is supporting SciDAC outreach, training and research objectives, another aim is to serve as an HPC information clearinghouse helping foster communication within the HPC community. Additionally, the Center will gather data to better understand the needs of the HPC community, as well as potential members of that community, and to identify workshops, summer schools, institutes and research topics to meet those needs.
When technology companies see a possible role for their new architectures or methods in the scientific computing space, the Outreach Center provides a conduit between technology producers and scientific applications within SciDAC. The Center recently worked with a database appliance company to find researchers who could make use of their technology. That dialogue started out as a request to help identify scientific computational workloads with a good match with the company’s technology and to set up meetings as a way of getting connected with the scientific computing community.
Initial point of contact
In establishing the Outreach Center, DOE’s Office of Advanced Scientific Computing Research clearly indicated that it should serve as a point of contact for potential HPC users from industry who are looking to increase the contributions of computing within their organizations. The Center can help answer queries such as• What HPC resources are available?
• How do potential users request allocations on the system?
• How can the performance of applications be improved?
• What tools are out there to develop and improve applications?
• How do I take advantage of these resources?
The Outreach Center is located at DOE’s National Energy Research Scientific Computing (NERSC) Center at LBNL in California. Established in 1974, NERSC has a proven track record of helping researchers make the most of the center’s HPC resources. Many of NERSC’s 2,900 users have received help in scaling their codes up from running on dozens of processors to thousands of CPUs, thereby enabling them to run more detailed simulations of increasingly complex problems in less time.
 click to enlarge Pellet injection is a mechanism for the refueling of burning plasma reactors. The physical processes are inherently multiscale which has prevented effective simulations in the past. Block-structured AMR allows researchers to efficiently simulate pellet injection and investigate in detail the fueling and mass redistribution processes in a tokamak. Top Left: Block-structured adaptive curvilinear meshes to resolve the pellet. Top Right: Volume rendering of pressure shows hot spot due to electron heating (H-W Shen & S. Martin, Ohio State University). Bottom: Density slices show effective resolution of the pellet (S. Ahrens, ORNL/VACET). |
Since it was launched in 2001, a hallmark of the SciDAC Program has been to build stronger connections between different disciplines. For example, the projects to develop scientific applications in areas ranging from molecular chemistry to climate change to astrophysics include scientists and engineers, applied mathematicians and computer scientists. In 2006, SciDAC also established a number of centers for enabling technologies and university-based institutes with the goal of making the research benefits of SciDAC available to the broader research community.
Bridging gaps
The Center has a two-pronged approach to making SciDAC and HPC more accessible to new communities. The first is a personal approach to outreach by augmenting conferences and conducting training events to include SciDAC-related topics. The second is an array of IT offerings that help SciDAC projects organize and transfer their technologies to new audiences. Web hosting, software packaging and distribution, and collaborative development environments are among the services that the Center provides to SciDAC projects. All of these services help to bridge the gap between SciDAC efforts and HPC needs that exist outside of the research areas well known to the computer scientists, applied mathematicians and application specialists present in the many aspects of DOE computing. The Outreach Center serves to find useful applications of SciDAC-supported technologies in industry, government and academia.Last June, the Center conducted a day of six tutorials in wide-ranging SciDAC-supported areas in conjunction with the SciDAC 07 Annual Meeting in Boston. Nearly 100 students from the surrounding universities attended the workshop to gain hands-on and in-person training in HPC. The tutorials provided direct access to both DOE’s HPC experts and trial use of DOE computational resources. Several students provided evaluations of the tutorials, indicating this was their first-ever hands-on access to DOE parallel computing resources.
 click to enlarge Lock-exchange over a Gaussian sill. Top plot: density. Middle plot: vorticity. Bottom plot: a passively advected scalar initialized to equal the depth. This computation was done using a 128×8 grid at the coarsest level and three additional levels of AMR (refinement ratio 4:1), yielding an effective resolution of 8192×512 at the finest level. Courtesy of Michael F. Barad, Environmental Fluid Mechanics Laboratory, Stanford University. |
In one tutorial, the students had a chance to register their local computing resources with the Open Science Grid, a jointly funded DOE and National Science Foundation (NSF) effort to deliver computing and data resources to wide-ranging audiences. Such shared time at the keyboard between the expert and novice goes a long way in bridging gaps to new applications of HPC.
While some tutorials had a practical “installfest” direction, other sessions were more academic in nature, providing, for instance, instruction in the solution of partial differential equations using modern computational approaches. Tutorials in data management, parallel IO and scientific visualization also were conducted.
The Center has had a particularly strong interaction with DOE’s Computational Science Graduate Fellowship program which targets harnessing the best and brightest students in the realm of computational science with mission goals of the Office of Advanced Scientific Computing Research.
Matching new software with potential users
In addition to education-based activities directed at academics and tomorrow’s crop of computational scientists, the Outreach Center has a software engineering side as well, aiming to assist in the delivery of SciDAC-supported technologies through improved software practices, since software that is well-designed, well-tested, and well-packaged has a greater chance of effectively delivering results to new audiences. The Center provides assistance in making the software that represents the research goals of SciDAC centers and institutes more widely accessible. Tools developed by APDEC are being used to advance research in areas such as underwater flows, fusion energy, combustion and astrophysics. click to enlarge A heliopause instability in the presence of charge exchange with interstellar neutrals (to appear in the Astrophysical Journal). Colormap: the logarithm of the plasma temperature in the heliosphere. The computation was done using a 272×128 grid at the coarsest level and five additional levels of AMR (refinement ratio 2:1), yielding an effective resolution of 8704×4096 at the finest level. Courtesy of: Sergey N. Borovikov, Nikolai V. Pogorelov, Igor A. Kryukov, Gary P. Zank, Institute of Geophysics and Planetary Physics, University of California, Riverside. |
An example of software which has successfully made the transition from the applied math space to ubiquitous availability on all computing platforms includes the Basic Linear Algebra Subroutines (BLAS). BLAS routines are widely used in computing projects that include DOE as well as industry and academic research objectives. Software engineering, such as that provided by the Outreach Center makes this possible. Other software, such as LAPACK and ScaLAPACK, approach a similar level of general use. There are many other technologies ranging from adaptive mesh refinement to graph partitioning that are in transition to wider audiences.
A bi-directional flow
However, the flow of information through the SciDAC Outreach Center is not a one-way street. The center also serves as a gateway for new technologies and ideas to come to the SciDAC research community. One example of brokering such technology is collaboration around the offerings of Netezza in the context of data analytics. In this collaboration, a technology provider from industry contacted the Center to determine if there might be overlap between their directions and the needs of the HPC community. Netezza manufactures a data analytics appliance that the Outreach Center made available to scientific application partnerships in SciDAC. This resource has become a testbed for new approaches to data analysis. Among other projects, the analysis of combustion simulation data has benefited from access to resources made available by new outward-facing approaches to SciDAC outreach.Conclusion
Outreach is a multifaceted objective. Making both resources and needs that exist within the scope of SciDAC research available to new audiences is crucial to the advancement of the scientific discovery that SciDAC was established to deliver. Whether it is bringing new talent to HPC, delivering HPC resources to new audiences, or incorporating new technologies into DOE’s HPC portfolio, the SciDAC Outreach Center is there as a broker between resources and needs to help get computing done.Reference
1. J. B. Bell, M. S. Day, J. F. Grcar, M. J. Lijewski, J. F. Driscoll and S. F. Filatyev, “Numerical Simulation of a Laboratory-Scale Turbulent Slot Flame”, Proc. Combust. Inst. 31 1299-1307 (2007).David Skinner leads the SciDAC Outreach Center. He may be contacted at [email protected].
Acronyms
BLAS Basic Linear Algebra Subroutines | CCSE Center for Computational Sciences and Engineering | DOE U.S. Department of Energy | LBNL Lawrence Berkeley National Laboratory | NERSC U.S. National Energy Research Scientific Computing | NSF U.S. National Science Foundation | ORNL Oak Ridge National Laboratory | SciDAC Scientific Discovery through Advanced Computing Program | VACET Visualization and Analytics Center for Enabling Technologies
