***Wednesday December 8th***

9:00 am - 9:45 am

HPC in the Multicore Era - Challenges and Opportunities
Dr. David Barkai
HPC Computational Architect
Intel Corporation

The application of "Moore's Law" still provides us with doubling the performance every generation through higher density of transistors on the chip, but also with the help of architectural innovation. What is a new phenomenon is that since the introduction of multicore the potential performance increase will not occur without assistance from the software and applications developers community. The challenges arise from the ever increasing level of concurrency, and the innovation required to maintain an adequate compute-bandwidth-latency balance. This talk will cover the current state of affairs in HPC, the hardware and software challenges, some approaches being studied to resolve them - new approaches to memory and IO systems, examination of more complex programming models (i.e., heterogeneous and hybrid); and the scientific discovery opportunities that will open up when those barriers are overcome.

Speaker Bio:
Dr. David Barkai is an HPC computational architect for Intel Corporation, involved in interfacing between the HPC users' community and Intel. He also held a number of positions within Intel research labs involving peer-to-peer and investigations of the impact of emerging technologies on society. Before joining Intel in 1996, David worked for over 20 years in the field of scientific and engineering supercomputing for Control Data Corporation, Floating Point Systems, Cray Research Inc., Supercomputer Systems Inc., and NASA Ames Research Center. David received his B.Sc. in physics and mathematics from the Hebrew University in Jerusalem and a Ph.D. in theoretical physics from Imperial College of Science and technology, London University, in the UK. Over the years David has published in various forums on subjects in physics, numerical methods, and computer applications and architectures. He authored the book "Peer-to-Peer Computing: Technologies for Sharing and Collaborating on the Net" (Intel Press, 2001)
(back to top)

1:00 pm - 2:00 PM

MPI-IO Tuning for Application Parallel I/O
Bill Loewe
Panasas

This session focuses on MPI-IO starting from the application perspective, covering advantages/disadvantages, setup of independent/collective I/O, contiguous/discontiguous datatypes, and MPI-IO hints. This sets the stage for looking at how MPI-IO then interacts with the underlying file system as well as the various I/O designs the application can take advantage of to improve performance.
(back to top)

2:00 pm - 3:15 pm

Hands-on: Build your own Supercomputer (part I)
Hosted by the HPC Advisory Council

The HPC Advisory Council (www.hpcadvisorycouncil.com) is a high-performance computing educational and outreach center, consisting of more than 160 organizations. The HPC Advisory Council includes five special interest subgroups – HPC|Works, HPC|Scale, HPC|Storage and HPC|Cloud, and HPC|GPU and operates HPC educational centers around the world. The hands-on session will provide a unique opportunity to learn and experience building a leading-edge HPC cluster, how to configure, troubleshoot, optimize and manage it and will also include an overview on new technologies, such as GPUDirect. The session will be extremely beneficial for HPC users (new or experienced), for people who manage or plan to manage HPC clusters, and to anyone who wants to build their own supercomputer.
(back to top)

3:30 pm - 5:00 pm

Hands-on: Build your own Supercomputer (part II)
Hosted by the HPC Advisory Council

The HPC Advisory Council (www.hpcadvisorycouncil.com) is a high-performance computing educational and outreach center, consisting of more than 160 organizations. The HPC Advisory Council includes five special interest subgroups – HPC|Works, HPC|Scale, HPC|Storage and HPC|Cloud, and HPC|GPU and operates HPC educational centers around the world. The hands-on session will provide a unique opportunity to learn and experience building a leading-edge HPC cluster, how to configure, troubleshoot, optimize and manage it and will also include an overview on new technologies, such as GPUDirect. The session will be extremely beneficial for HPC users (new or experienced), for people who manage or plan to manage HPC clusters, and to anyone who wants to build their own supercomputer.
(back to top)

5:00 pm - 5:30 pm

Finding New Users for HPC Clusters
Rich Altmaier
Director of Marketing and Business Development
Intel Cluster Ready, Software and Services Group, Intel Corporation

Intel has developed a cluster platform software architecture to help expand the users of HPC cluster systems, making it easier for new HPC users to select, deploy, and use clusters. The next million users will not be the same as the first 100k.

Speaker Bio:
Rich Altmaier joined Intel in Apr 2008 to improve computing methods in the industry. He is now part of the Intel Cluster Ready program. He has been focused on high performance computing systems for his entire career. Rich was Director, and then VP of Engineering, for SGI for 15 years, managing the development of Operating System software, including seven supercomputer systems (Challenge through Altix). The last six years have been in the Linux environment, with his team contributing to the growth of Linux to serve high performance applications.
(back to top)

***Thursday December 9th***

8:45 am - 9:45 am

How to Succeed in MPI Without Really Trying
Dr. Jeff Squyres
MPI Architect at Cisco Systems, Inc.

* Does writing parallel code humble your soul?
* Do you frequently feel, "Well, here goes nothing..." when you run your applications?
* Have you ever uttered, "My code works at 32 nodes, but fails at 64"?
* Do you hope and pray that Santa brings you bug-free MPI code for Christmas?

If you answered "yes" to any of the above questions, KNOW THAT YOU ARENOT ALONE! Come hear some hard-won knowledge of common mistakes to avoid when writing portable parallel MPI applications.
(back to top)

10:00 am - 11:00 am

The road to Exascale computing
Gilad Shainer
HPC Advisory Council

PetaScale and Exascale systems will span tens-of-thousands of nodes, all connected together via high-speed connectivity solutions. With the growing size of clusters and CPU/GPU cores per cluster node, the interconnect needs to provide not only the highest throughput and lowest latency, but to be able to offload the processing units (CPUs, GPUs) from the communications work in order to deliver the desired efficiency and scalability. The session will review the latest acceleration technologies over InfiniBand such as CORE-Direct, SHMEM, GPUDirect as well as the roadmap for the next generation InfiniBand solution.
(back to top)

11:00 am - 12:00 pm

Hybrid Vigor: Using heterogeneous HPC to accelerate chemical biology
Imran Haque
Pande lab, Small molecule design
Graduate student, Computer Science
Stanford University

Fifteen years ago, the advent of modern high-throughput sequencing revolutionized computational genetics with a flood of data. Today, high-throughput biochemical assays promise to make biochemistry the next data-rich domain for machine learning. However, existing computational methods, built for small analyses of about 1,000 molecules, do not scale to emerging multi-million molecule datasets. I will describe how new algorithms and new hardware (GPUs) allow us to cross a 10,000-fold scalability barrier to do large-scale biochemical machine learning.
(back to top)

Millisecond timescale, atomistic simulations reveal that protein folded states are kinetic hubs
Greg Bowman
Postdoctoral Fellow
Pande lab
Stanford University

Understanding protein folding is a classic grand challenge in molecular biophysics; a solution for which could have immediate medical benefits, particularly for protein misfolding diseases like Alzheimer's.Molecular Dynamics (MD) simulations have the potential to provide quantitative models of protein folding but, unfortunately, this potential has yet to be fully realized due to the need to capture long-timescale transitions at atomic resolution.We have developed a network approach to combining many simulations into a single statistical model capable of capturing events on timescales longer than any individual simulation.We can also exploit these network models to direct sampling to where it is needed most, allowing for tremendous computational efficiency.Using these models, we can now predict the structure, thermodynamics, and kinetics of proteins with up to 80 amino acids and 10 millisecond folding times.From these models, we have discovered that protein folded states are actually kinetic hubs, rather than being kinetically isolated from unfolded conformations, as was previously thought.
(back to top

1:00 pm - 2:00 pm

Optimization of a Formula 1 Wheel Assembly using CFD and Mesh Morphing Technology on Certainty
John Axerio-Cilies
PhD Candidate
Mechanical Engineering
Stanford University

The geometrical uncertainty associated with a rotating rubber tire can greatly influence the optimal race car setup for a particular track. We quantify this uncertainty by creating up to 20,000 unique 3D wheel configurations that span an input space of multiple uncertain and design parameters. The aerodynamic characteristics of each wheel configuration, consisting of 30 million cells, is solved numerically using Computational Fluid Dynamics (CFD) on Certainty. Every configuration is slightly modified by morphing the mesh locally depending on the multi-objective optimization algorithm.
(back to top)

High-fidelity LES for supersonic rectangular jet noise prediction
Dr. Joseph W. Nichols
Research Associate
Center for Turbulence Research

Engine exhaust noise is a critical design consideration for next generation civil and high-performance aircraft. As communities continue to expand, close proximity to airports seems inevitable, resulting in increasingly stringent noise regulations. Also, jet noise is a particular problem for aircraft carrier flight deck crews, as even the most advanced hearing protection devices cannot offer adequate protection from the 150 dB noise produced by present-day aircraft at take-off. To aid in the reduction of jet noise at its source, our project focuses on assessing the capability of a fully unstructured large-eddy simulation code (CharLES) coupled with a Ffowcs Williams and Hawkings solver to accurately predict the noise produced by supersonic, turbulent jets issuing from shaped nozzles. In particular, we consider a rectangular nozzle of aspect ratio 4:1, with the same geometry as one fabricated for upcoming laboratory experiments. For this nozzle, ee compare noise from both isothermal and heated operating conditions. In order to resolve as many scales as possible, 86 million control volumes are incorporated into the mesh. Results and scalability statistics from both the NSF MRI-2 Certainty cluster and a Cray XE6 system recently installed at the ERDC supercomputing center, where the code has been run on as many as 20,000 cpus concurrently.

This research is supported by NASA grant no. NNX07AC94A
(back to top)

Large Eddy Simulation of Supersonic Combustion
Amirreza Saghafian
PhD Candidate
Flow Physics and Computational Engineering Group
Department of Mechanical Engineering, Stanford University

A combustion model based on a Flamelet/Progress Variable approach for high-speed flows is introduced. In the proposed formulation, the temperature is computed from the transported total energy and tabulated species mass fractions. The combustion is thus modeled by 3 additional scalar equations and a chemistry table that is computed in a pre-processing step. This approach is very efficient and allows the use of complex chemical mechanisms. An approximation is also introduced to eliminate costly iteration steps during the temperature calculation. To better account for compressibility effects, the source term for the progress variable is rescaled with the pressure. The model is tested in LES computations of a hydrogen jet in a supersonic transverse flow. Comparison with experimental measurements shows good agreement.
(back to top)

 

2:00 pm - 2:20 pm

Computer simulation of the structural changes of the antimicrobial peptide cecropin near a bacterial cell inner membrane
Amirali Kia
PhD Candidate
Stanford University

Recent developments in anti-bacterial polymeric coating technology allows inclusion of antimicrobial peptides on a surface coating. Antimicrobial peptides are known to attack and kill microbial cells by destroying their cell membrane. The size, sequence and structure of the antimicrobial peptide play significant role in its activity. We studied the structural changes of the antimicrobial peptide cecropin in bulk water and in the vicinity of bacterial-like membranes using extensive molecular dynamic simulations. We also computed the folding - unfolding free energy of the structure in both environments using Adaptive Biasing Force (ABF) method, developed in our group. A comprehensive study was performed to estimate the modeling and statistical errors in computing the free energy.
(back to top)

2:20 pm - 3:00 pm

Airframe Noise Prediction using Large Eddy Simulation
Chris Yu and Will Wolf
PhD Candidates
Aeronautics & Astronautics
Stanford University

Sustaining the continued growth of civil aviation is critically dependent on limiting its environmental impact. Noise regulations have become incrementally more stringent as air traffic has increased and will likely continue to do so in the future. As continued improvements in jet engine noise are made, airframe noise becomes a significant portion of the total aircraft noise signature, especially in landing approach. Since airframe noise sources, primarily high lift devices (flaps and slats) and landing gear, are typically geometrically complicated the cost of simulating these features can be prohibitive. Furthermore, accurate far-field sound calculations are extremely computationally expensive due to the large range of spatial and temporal scales that must be resolved. We apply Large Eddy Simulation to simplified geometries, inline tandem cylinders and a NACA 0012 airfoil, to study the generation of near field noise sources. These results will be used together with a fast acoustic analogy solver to compute far field noise.
(back to top)

A Parametric Investigation of Oblique Shock / Turbulent Boundary Layer Interaction
Brandon E. Morgan
Ph.D. Candidate
Stanford University
Aeronautics & Astronautics

Historically, high-fidelity large-eddy and direct numerical simulations (LES, DNS) have been considered too computationally expensive for use in wide-reaching parametric investigations. With recent advances in high performance computing, however, such feats are becoming more realizable. In the present investigation, large-eddy simulation (LES) of an oblique shock impinging on a supersonic turbulent boundary layer (M∞ = 2.28, ? = 6.5-9.5°, Reθ = 1500, 2300, 4800) is carried out with a high-order compact differencing scheme using localized artificial diffusivity (LAD) for shock capturing. Solution sensitivity is then investigated with regards to mesh resolution, domain size, Reynolds number, and wedge angle. Progressive mesh refinement and comparison with the literature are used to establish confidence in solution quality. It is found that the separation bubble is not significantly affected by increasing the spanwise domain beyond 3δ or by increasing the streamwise domain beyond 5δ past the shock impingement point. Additionally, the size of the separation bubble, while under-predicted with respect to experiment, does not appear to be significantly affected by Reynolds number (over the range considered). This leads us to speculate that the commonly observed discrepancy between simulation and experiment – which previously had been explained as a result of difference in Reynolds number between simulation and experiment – may instead be due to some three-dimensional confinement effect in the experiment. Finally, through analysis of the spectral content of the wall pressure signal in the separation bubble, the expected low-frequency motion is identified with a time scale ~O(100δ/u∞).
(back to top)

3:15 pm - 3:45 pm

Clean Energy: The Role of Computational Modeling of Catalysts
Christopher O'Grady, Ph.D.
SLAC/Stanford SUNCAT center
(SUstainable eNergy through CATalysis)

The negative effects of an increasing world population are becoming more apparent. The availability of clean/cheap energy is one of the most significant problems. The SUNCAT center led by Dr. Jens Norskov was created in July 2010 at SLAC/Stanford in order to try to computationally design catalysts important for clean/cheap energy (e.g. artificial photosynthesis, fuel cells). The science behind this will be discussed, as well as experiences with the computational tools used to do the associated modelling of catalytic systems.
(back to top)

3:45 pm - 4:15 pm

Advanced use of HPC Resources: Tips and tricks for clusters ranging from Stanford to National Lab resources
Presented by Users of Compute Resources at Stanford

This is a not-to-miss session. We have a number of users that will share information ranging from advanced use of resource managers, experiences of running on BlueGene to I/O and more.
(back to top)

4:15 pm - 4:45 pm

Architecting High Performance Computing Systems for Fault Tolerance and Reliability
John Fragalla
HPC Senior Technologist
Dell Inc.

As a High Performance Computing (HPC) Senior Technologist, John Fragalla brings global expertise to Dell’s HPC team. He is and has been a leader on many large HPC opportunities globally, providing technical advisement on future product development and direction within Dell and the industry to architect solutions that meets customer requirements, leads training sessions to increase Dell’s HPC knowledge, and is one of the HPC solution development leads. John works with engineering on product definition by providing global customer requirements to influence product specification on new product direction for the HPC community. John is considered one of the primary HPC resources, working with the global HPC community, and is viewed by customers, external partners, and Dell employees as a leader in his field. John represents Dell at many customer events and conferences, and recently published a technical report on the use of InfiniBand in HPC.

While solving many customer requirements and developing end-to-end HPC architectures globally, John expertise includes, but not limited to, HPC storage and data management, x64 system designs, interconnects and I/O subsystems, power and cooling, and GPGPU computing. Previously in joining Dell’s HPC team, John Fragalla was a Global HPC Principal Engineer at Sun Microsystem’s and a software developer at Johnson and Johnson. As well as many years of practical HPC field experience, John holds a Bachelor of Science in Computer Engineering from the University of California, Irvine and a Master of Science in Electrical Engineering, specializing in Computer Architecture and Parallel Computing, from the University of Southern California.
(back to top)