PSC Live! at SC07
Tuesday, November 13th
A new scalable approach for solution of large computational biomechanics problems on modern supercomputers will be presented. Specifically, a 3D simulation of blood flow in the human cranial arterial tree using the spectral/hp element method is considered. Spatial discretization of the 3D computational domain of 65 cranial arteries results in a half-million tetrahedral elements and over a billion degrees of freedom. This talk will focus on multi-layer decomposition of computational domain and multi-layer decomposition of the global communicator implemented in Nektar-g2, a parallel numerical solver developed at Brown University.
PSC has developed a prototype distributed file system infrastructure that vastly accelerates aggregated write bandwidth on large compute platforms. Write bandwidth, more than read bandwidth, is the dominant bottleneck in HPC I/O scenarios due to writing checkpoint data, visualization data and post-processing (multi-stage) data. We have prototyped a scalable solution on the Cray XT3 compute platform that will be directly applicable to future petascale compute platforms having of order 106 cores. Our design emphasizes high-efficiency scalability, low-cost commodity components, lightweight software layers, end-to-end parallelism, client-side caching and software parity, and a unique model of load-balancing outgoing I/O onto high-speed intermediate storage followed by asynchronous reconstruction to a 3rd-party parallel file system. The absence of a central metadata service further reduces latency, allowing for the maximum reliable performance per unit cost for petascale systems.
Distributed memory agent-based models (ABM) have been used in recent years to predict the spread of disease in human populations. In general, ABMs have simplifying assumptions that enable the model to run on small clusters (for local-level models) or supercomputers (for national scale models). However, more detailed local-level discrete-event models (e.g. EpiSims) have demonstrated the added utility of detailed local ABMs to understand the dynamics and constraints of public health interventions during infectious disease outbreaks. To enable the needed simulations we utilize the vast capacity of the National NSF TeraGrid resources with a grid-aware implementation of EpiSims using MPICH-G2 and a new distributed memory parallel implementation that exploits computational locality.
Wednesday, November 14th
The goal of this new program is to bring innovative science tutorials into secondary school classrooms, focusing on integrative computational biology, physiology, and biophysics. CMIST modules include high quality, biologically realistic 3-D animations produced with cutting-edge simulation and visualization software developed by NRBSC. Subject material is chosen to highlight critical concepts that are difficult to convey with static textbook images and words; for example, cellular transport of molecules by diffusion and osmosis, or the structure and function of enzymes in cells. Thus, CMIST modules bring a visually-engaging but highly accurate view of biological and medical function into the classroom, providing a transformational means to improve student understanding and retention of complex multidisciplinary material.
Cardiovascular disease is the cause of a large number of deaths in the developed world. Cerebral blood flow behavior plays a crucial role in the understanding, diagnosis and treatment of the disease. The GENIUS project aims to model and characterize large-scale patient-specific cerebral blood flow in clinically relevant run times by means of leading-edge computing systems, remote visualization, and Grid technologies. This talk will present an overview of the approaches and activities, focusing on the development and optimization of the HemeLB fluid solver and its visualization system to use PSC's Cray XT3 and UK NGS HECTOR XT4 systems.
Numerous studies have shown the value of experiential learning on mastery of complex topics. PSC proposes to extend this principle to education in STEM (Science, Technology, Engineering and Mathematics) by developing and deploying STAKEs---Science and Technology Active Knowledge Experiences. STAKEs are opportunities for students to explore and experience the impact of science and technology in active rather than passive ways. Based on the CI Jumpstart materials developed to introduce high school students to cyberinfrastructure, and other hands-on experiences distributed at local and national outreach events, PSC has established a suite of activities to engage students and their teachers in fundamental concepts critical to computational thinking.
Thursday, November 15th
The PyProf profiling tool provides loop-level counts of memory accesses and floating-point operations suitable for inclusion in detailed analytic performance models used to predict the performance of future architectures. PyProf provides valuable functionality complementary to that of tools such as TAU. PyProf currently supports both Fortran and C, and has been applied to MILC, PKDGRAV, and other applications of interest.
An efficient technique for the real-time visualization of numerical solutions of a large scale problem is examined. The method, developed by a team from Brown University and the Pittsburgh Supercomputing Center, is designed for computations performed at remote supercomputers and visualization on local or remote computers. It was implemented in Nektar, a parallel numerical solver based on high order spectral/hp element method, coupled with a modified version of ParaView. The efficiency of transferring data from the solver to ParaView derives from Portals Direct I/O (PDIO) developed at PSC. PDIO routes traffic from compute nodes on the internal high-speed network to arbitrary hosts on the Wide Area Network. Developed in collaboration with the PSC visualization group, a new data file format and reciprocal modification of ParaView resulted in more than an 80% reduction in the size of visualization data.
PSC Education Program Schedule
Birds of a Feather Session (Technical Program)Session: Career Paths: High Performance Computing Education and Training
Leaders: Laura McGinnis, PSC; Mary Ann Leung, University of Washington; Brad Armosky, TACC
Date: Tuesday, November 13
Time: 12:15pm - 1:15pm
Abstract: In the next 5 years, the amount of computing resource cycles available for scientific research will increase significantly. Exploiting this explosion in computing power poses opportunities and challenges. The early adopters of these systems are in our classrooms today. This year’s freshman will graduate mere months before NSF’s target date for production petascale computing. Graduate students are already in place, learning from pioneers in computational science. These are the users who will advance their fields of science into the next generation of computation.
The NSF TeraGrid community is working to prepare the next generation of users for the next generation of resources. This BoF invites computational scientists from all disciplines and agencies to join in a discussion of HPC career development pathways. Topics will include pedagogies, curricula, and training and education resources that are available or need to be developed to meet the needs of future computational scientists.
Education ProgramParallel Session: HPC University Resources
Speakers: Laura McGinnis, PSC; Mary Ann Leung, University of Washington
Date: Tuesday, November 13
Abstract: In the next 5 years, the scientific community will experience an unprecedented explosion in HPC resources that will dwarf current offerings. By conservative estimates, the NSF alone will put over 10 petaflops in additional cycles in the hands of the newest generation of computational scientists by 2011. By comparison, the total capacity of the entire top500.org list (Nov ’06) is only 3.5 petaflops. Exploiting this explosion in computing power poses significant opportunities and challenges to computational scientists. The early adopters for these systems are in our classrooms today. The incoming freshman university class will graduate in 2011, mere months before NSF’s target date for production petascale (Track 1) computing. Graduate students are already in place, learning from pioneers in computational science. These are the users who will advance their fields of science into the next generation of computation. The NSF’s TeraGrid program is investigating successful paths to creating qualified, effective HPC professionals, capable of exploiting current super-terascale and upcoming petascale technologies for the advancement of scientific research, through a new initiative: HPC University. Gaps in training materials and delivery methods are being identified, and recommendations for filling these gaps will be presented at this overview presentation. Feedback from the HPC community is needed to advance this initiative and build it into a successful resource suite which will address the future needs of HPC.Parallel Session: CMIST
Speakers: Pallavi Ishwad, Jenda Domaracki, and Joel R. Stiles, PSC
Date: Tuesday, November 13
Abstract: The National Resource for Biomedical Supercomputing (www. nrbsc.org) and the Pittsburgh Supercomputing Center (www.psc.edu) are developing innovative Computational Modules In Science Teaching (CMIST) for use in secondary science classrooms. In contrast to many other teaching tools, CMIST modules are produced with highly realistic modeling and simulations based on software such as MCell and DReAMM (www.mcell.psc.edu), and are packaged with highly engaging, visually appealing, and scientifically accurate animations to convey critical concepts and examples. A pilot module entitled “Molecular Transport in Cells” presents important principles of osmosis and diffusion with realistic 3-D biological examples extrapolated from classic static textbook pictures, and is distributed as ready to use DVDs. The DVDs include a complete presentation, lecture slides, animations, lesson plan aligned to National and State Standards, and worksheets and answer keys. Thus CMIST modules are intended to be used “out of the box” and can be “dropped into” a biology, chemistry, physics, computer science, and/or math class period. Our approach to CMIST development has been heavily influenced by in-house experience with high school, undergraduate, and graduate teaching, and regional high school science teachers have already provided invaluable feedback through a recent series of NRBSC and PSC workshops. Based on this initial success, we are now introducing web-based distribution of CMIST materials, and are continuing development of additional CMIST modules for a variety of topics such as enzyme reactions, cellular metabolism, synaptic physiology, meteorology, geophysics, and astronomy. In this session we will briefly present our approach to CMIST module development, while focusing mainly on actual examples of CMIST content and use in the classroom.
Education BoothCI Jumpstart
Speaker: Laura McGinnis
Date: Wednesday, November 14
Abstract: CI Jumpstart is a presentation designed to be delivered to high school and undergraduate students, to introduce them to computational science and high performance computing. Originally presented as a 2 hour workshop, CI Jumpstart provides background material and encourages students to consider including scientific computing as an aspect of their science fair projects. It also serves as a STAKE (Science and Technology Active Knowledge Experience) for supporting STEM (Science, Technology, Engineering, and Math) activities. The CI Jumpstart presentation offered in the SC07 Education Booth will include the current CI Jumpstart slide set and an overview of issues to be considered when preparing a similar opportunity for students. Feedback from attendees will help tune and tone this material, to make it more accessible and effective in introducing students to this critical discipline and the opportunities it can open for them.Student Program: Internships at PSC and Beyond
Speaker: Laura McGinnis, PSC
Date: Wednesday November 14
Abstract: Post-secondary and corporate programs in science and engineering have traditionally made good use of internship opportunities to identify, mentor and develop students in their specialties. Computer and computational sciences, especially in academia and research laboratories may not be making as effective use of internship opportunities. Pittsburgh Supercomputing Center has had a long tradition of engaging undergraduate students from local universities to participate in day to day operations and research activities with a great rate of success. This session will present the PSC internship program, from both the students’ and Center’s perspective, including examples of the work that PSC students are doing to advance scientific computing as well as resource support. We will provide an opportunity for students, supervisors, and other interested people to talk about internship opportunities and offering pre-service opportunities to high school and college students. Discussions will focus on sharing successful strategies for capitalizing on the enthusiasm and availability of the next wave of scientists and technologists.