November 11, 2005
In this exclusive interview, HPCwire discussed recent HPC activities at the Pittsburgh Supercomputing Center with Nick Nystrom, PSC's director of strategic applications and a research physicist at Carnegie Mellon University. PSC is an NSF center and one of the premier facilities for computational science in the U.S.
HPCwire: PSC historically has been the first, or one of the first, to receive new HPC systems that major vendors develop. From a scientist's standpoint, what are the advantages of gaining access to systems this early on?
Nystrom: Accessing more powerful capability-class systems allows scientists to increasingly address the problems they really want to solve, rather than simplifications. One can go to realistic resolution and model all relevant degrees of freedom. This translates into a better return on investment. R&D cost is high, so rapidly translating the investment required to develop a new system into results of scientific, economic, and national importance ideally leverages that investment.
HPCwire: Big Ben, the PSC's Cray XT3, went into production on October 1. Can you provide some examples of scientific breakthroughs made possible through that resource?
Nystrom: Actually, breakthroughs began well before production. In April 2005, PSC scientists Troy Wymore and Shawn Brown performed a series of path integral quantum mechanical/molecular mechanics runs to elucidate the molecular basis of two different metabolic diseases, Hyperprolinemia Type II and Sjogren-Larsson Syndrome.
More recently, Yang Wang, in collaboration with Oak Ridge National Laboratory and Florida Atlantic University, performed a record-breaking electronic structure calculation on an iron nanoparticle embedded in an iron-aluminum binary compound. This 16,000-atom calculation required 1,600 Cray XT3 processors using LSMS 2.0, which Yang recently extended to facilitate designing new materials with specific magnetic properties.
HPCwire: How does Big Ben fit as an NSF TeraGrid resource?
Nystrom: Big Ben is PSC's newest contribution to the TeraGrid. The XT3's SeaStar interconnect, which features very high bandwidth and which at PSC is configured as a 3D torus, is ideal for challenging simulations that require thousands of processors. However, that's not to say that it's restricted to standalone, tightly coupled applications. For example, Nathan Stone, in PSC's Advanced Systems group, developed a library to allow I/O over the SeaStar's low-level Portals communications protocol, which is now being used by Prof. Paul Woodward to interactively steer very large CFD calculations. Paul and other researchers at the University of Minnesota can initiate calculations on Big Ben, visualize in real time the results on their power wall, and adjust parameters of the calculation as it executes. This interactivity will greatly reduce the time to solution for important problems in turbulence.
HPCwire: How much impact do advances in HPC technology have on advances in science? How direct is the relationship between the two?
Nystrom: The impact of HPC on science can be tremendous for a variety of reasons. Large-scale computations can probe phenomena that are observationally inaccessible or that would be prohibitively expensive. A nice example of that is work by the Quake group at Carnegie Mellon University. Their simulations of earthquakes in the Los Angeles basin require detailed knowledge of subsurface geology, which they infer through a computationally demanding inverse procedure. New end-to-end runs on the XT3, including everything from meshing through visualization, will allow a groundbreaking 2Hz simulation on over 10 billion elements. Going to those levels of resolution is necessary to obtain better predictions of ground motion, which in turn can guide construction planning and disaster preparedness.
Similarly, work performed by Prof. Klaus Schulten and Emad Tajkorshid of the University of Illinois on PSC's systems elucidated the passage of water, but not protons, through aquaporins, which are important membrane proteins. Their work complemented the experimental work for which Dr. Petre Agre received the 2003 Nobel prize in chemistry, and a visualization of their aquaporin simulation is available at the Nobel website.
HPCwire: I understand that LeMieux, PSC's big Alpha system, is still heavily utilized. How are people taking to Big Ben, your new 10-teraflop Cray XT3?
Nystrom: Interest has been quite high. Prior to Big Ben going into production, a number of research groups participated as "friendly users", gaining experience and porting their applications as we brought the system to readiness. Most of those have gone on to apply for production allocations, and others are transitioning from LeMieux.
Much of Big Ben's appeal is due to performance. Considering only clock speeds, Big Ben might be expected to be only 2.4 times as fast as LeMieux, processor-for-processor. But that doesn't take into account communications, which for many realistic applications are critically important. To illustrate that point, PSCC, a Parallelized Spectral Channel Code for simulating turbulent boundary layers, runs up to 10.5 times as fast on 512 XT3 processors as it does on 512 LeMieux processors. That improvement is due to the interconnect bandwidth.
HPCwire: What are some scientific problems or disciplines that will benefit most from sustained petaflop speed?
Nystrom: It's difficult to think of disciplines that would fail to benefit, given the desire to ask the hard questions and the resources to address them. Biophysics, engineering, materials science, ecology... Really, however, the greatest rewards will be obtained as simulations span disciplines and address ranges of scales. For example, from a molecular viewpoint, protein-enzyme interactions are extremely demanding computationally, with ab initio molecular dynamics calculations easily consuming hundreds of thousands of CPU hours. But those reactions occur in the context of cells, and modeling the system across that range, including reactions, kinetics, permeation and microphysiology will ultimately be required to understand processes such as cell metabolism and drug delivery. Such mesoscale modeling will require vast resources, and it is not unique to biology. It applies to most scientific domains.
HPCwire: Is HPC being used as aggressively as it should be in science?
Nystrom: Researchers using computational science today are obtaining excellent results and clearly demonstrating the role for HPC throughout science. However, we can do a better job of communicating to those who are not yet using HPC that hardware and software resources, together with expertise in using those resources, are available to help them. Also, there is an ongoing need for investment in scalable software and infrastructure. As we scaled our systems to thousands of processors, we met and overcame new challenges, and we'll have to do that again as we scale to petascale systems.
HPCwire: We often hear people talk about a shortage of people who have expertise in both their scientific discipline and computational science. Do you see this as a problem?
Nystrom: Such individuals exist, although due to specialization in scientific domains and increasing sophistication of software, people who are proficient in both certainly aren't common. Supercomputing centers play a vital role in that respect, acting as a bridge between domain experts and the resources needed to address the science. This need will continue to evolve as systems scale to the petaflop regime, with significant challenges in software design and implementation.
HPCwire: What can we expect to see from PSC at SC05 and beyond?
Nystrom: PSC will present exciting work representing a variety of interests in applications, networking, biomed, and systems. We've seen great scientific results on Big Ben, and a number of users will be present to demo and present their work first-hand.
PSC is intimately involved in SC05 networking, including supporting the Open InfiniBand (OpenIB) network for SCinet. Our networking group will also make presentations and perform demonstrations on the NPAD (Network Path and Application Diagnostics) project, which is funded by the NSF's "Strategic Technologies for the Internet" Initiative and addresses problems of network path delay inherent in transmitting data across wide area networks. They'll also show the HPN-SSH, a high-performance secure shell, which eliminates a known ssh bottleneck by allowing the flow control buffers to be defined at runtime.
PSC's Biomedical Initiative will be showing a variety of applications, including MCell for modeling stochastic microphysiology, DReAMM for model building and visualization, the PSC Volume Browser for volumetric visualization and analysis of massive datasets, and the Dynamo molecular simulation program for hybrid quantum mechanical/molecular mechanics simulations. In coming years, PSC will develop new web gateways for each of these applications, and workshops by the biomed group will address spatially realistic cell modeling, volumetric data analysis and visualization, and bioinformatics, especially targeting minority-serving institutions.
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?