A-Z Index | Phone Book | Careers

InTheLoop | 08.22.2011

August 22, 2011

Jay Srinivasan Named NERSC’s Computational Systems Group Lead

Jay Srinivasan has been selected as the Computational Systems Group Lead in the NERSC Systems Department. In this role, he will supervise the day-to-day operation of all of NERSC’s computer systems.

Jay has over 15 years of experience in high performance computing both as a user and administrator. Since joining NERSC in 2001, he has worked on all the large systems from NERSC-3 (IBM/SP2) to Hopper (Cray XE6) and was the system lead for the Jacquard system. Most recently, Jay was the team lead for the PDSF cluster that supports Nuclear Physics and High Energy Physics. Prior to NERSC, Jay worked at the Supercomputing Institute at the University of Minnesota, where he received his Ph.D. in Chemical Physics from the University of Minnesota.


Taking a Disruptive Approach to Exascale

Early in August the DOE Office of Advanced Scientific Computing Research (ASCR) held a workshop called “Exascale and Beyond: Gaps in Research, Gaps in our Thinking” that brought together luminaries from the world of high performance computing to discuss research and practical challenges at exascale. John Shalf of NERSC gave a presentation on “Exascale: Past, Present, Future.”

HPCwire highlights a few noteworthy presentations in this article. Detailed slides and other materials to further the conversation can be found here.


Farewell Reception for Juan Meza This Thursday

As you all know, Juan Meza, acting director of the Computational Research Division and head of CRD’s High Performance Computing Research Department, will be leaving us to take on his new role as Dean of Natural Sciences at UC Merced. Juan’s last day at Berkeley Lab will be this Friday, August 26.

Please join us in the Lab cafeteria for Juan’s farewell reception from 3:00 to 4:30 pm on Thursday, August 25. This will be an opportunity to thank Juan for his contributions to the Lab, both as HPCRD Department Head and most recently as Acting Division Director, to say farewell, and to wish him the very best at UC Merced. Light refreshments will be served.


ASCAC Meeting This Week

The DOE Office of Science’s Advanced Scientific Computing Advisory Committee (ASCAC) is meeting this Tuesday and Wednesday, August 23–24, in Rockville, MD. You can read the agenda here.


ESnet to Present Webcast on How to Use Globus Online

Got a lot of data to move around? Say, more than 100 Gigabytes? ESnet recommends that you attend this webcast at 11:00 am (Pacific time) Thursday, September 8, on how to use Globus Online, one of ESnet’s recommended file transfer services. The webcast will show you how to move data as you need it without having to become an IT expert, learn a new command vocabulary, or install software.

Globus Online Project Lead Steve Tuecke and ESnet Advanced Network Technologies Group Lead Brian Tierney will demonstrate the signup and transfer process, show some of the most frequently used Globus Online features and tips and tricks, and offer time for Q&A. Read more.


This Week’s Computing Sciences Seminars

Uncertainty Quantification Study Group: Verification in V&V and UQ
Monday, August 22, 10:00–11:00 am, 50B-4205
Dan Gunter, LBNL/CRD

The group will discuss the papers The T-Experiments: Errors in Scientific Software and Validation and Verification in Computational Fluid Dynamics (Chapters 1–3 only). Supplementary: Verification and Validation Benchmarks.

Single Particle Analysis via Manifold Embedding
Wednesday, August 24, 10:00–11:00 am, 50F-1647
Chunhong Yoon, University of Wisconsin–Milwaukee

The ultrafast pulses from x-ray free-electron lasers are of high enough intensity and of sufficiently short duration that individual single-shot diffraction patterns can be obtained from a single particle before significant damage occurs. This “diffraction before destruction” method may allow the determination of structures of proteins that cannot be grown into large enough crystals or are too radiation sensitive for high-resolution crystallography. The key challenges of this method are in obtaining a homogeneous set of particles and accurately determining the orientations of these snapshots. Manifold embedding techniques in machine learning possess great organizational ability in dealing with such large datasets. This talk will address some of the associated challenges in single particle analysis via manifold embedding.

Post-CMOS Strategy and Carbon Nanoelectronics
Friday, August 26, 11:00 am–12:00 pm, 3110 Etcheverry Hall, UC Berkeley
Chun-Yung Sung, IBM T.J. Watson Research Center

Information Technology (IT) industries currently have more than $200 billion in global sales and account for 30% of U.S. GDP and 50% of US economic growth. The unprecedented growth of the IT industry has largely been driven by the nonstop exponential increase in the performance of the CMOS-FET per unit area/dollar, which is enabled by the ability to continue scaling down CMOS transistor sizes and increasing functionality.

However, industry’s ability to scale transistors has become limited recently due to increasing leakage power and inability to reduce switching energy. Unavoidably, the fundamental limitations destine CMOS scaling to a conclusion at around 5-10 nm in 2020.

As CMOS shrinks closer to the point where it can’t get any smaller, an innovative new device and its architecture for the future logic switch becomes very urgent.

The post-CMOS device should show significant advantages in power, performance, density, and cost to enable the extension of the historical cost and performance trends for information technology. Amongst many promising options, the graphene device based on the unique electron transport characteristics has attracted a lot of attention due to their superior electrical and mechanical properties. Logic device based on graphene will have to be built on a new concept, one that takes advantage of the material's unique properties in a revolutionary architecture.

Moreover, if this new switch can be simultaneously dynamically reconfigured to perform multifunction logic operations must be very attractive.

Advanced Materials for Computing and Storage
Friday, August 26, 12:00–1:00 pm, OSF Room 238
Catherine Jenkins, LBNL/ALS

CMOS and related technologies will fail to keep pace with computing requirements by the end of the decade. This talk will muse on the design and characterization of materials for current and future problems of processing and storage, including but not limited to half-metallic ferro- and ferrimagnets for spintronics and topological insulators for quantum computing.

Searching for the Milli-Volt Switch
Friday, August 26, 2:00–3:00 pm, 390 Hearst Memorial Mining Bldg., UC Berkeley
Eli Yablonovitch, Director, Center for Energy Efficient Electronics Science, UC Berkeley

In contemplating the headlong rush toward miniaturization represented by Moore's Law, it is tempting to think only of the progression toward molecular sized components. There is a second aspect of Moore's Law that is sometimes overlooked. Because of miniaturization, the energy efficiency of information processing steadily improves. We anticipate that the energy required to process a single bit of information will eventually become as tiny as 1 electron volt per function, truly indeed a molecular sized energy.

Inevitably, most logic functions, including storage, readout, and other logical manipulations, eventually will be that efficient.

There is, however, one information-processing-function that bucks this trend. It is communication, especially over short distances. Our best projections of improvements in the short distance communication function show that it will still require hundreds of thousands of electron volts just to move one bit of information the tiny distance of only 10 micrometers.

Why this energy per bit discrepancy for communications? It is caused by the difference in voltage scale between the wires and the transistor switches. Transistors are thermally activated, leading to a required voltage >>kT/q. Wires are long, and they have a low impedance, allowing them to operate efficiently even at ~1 millivolt.

The challenge then is to replace transistors with a new low-voltage switch that is better matched to the wires. I will present some of the technical options for such a new switch that are being explored by the new NSF Science & Technology Center for Energy Efficient Electronics Science.


Link of the Week: To Err Is Primate

The question of how to best make decisions has fascinated humankind for centuries. For economists, the answer has always been relatively simple: making good decisions is a simple act of comparison shopping. A smart decision maker should start by listing all the possible choices for a given decision and then estimate the average payoff of each individual choice. Once the decision maker has all this information handy, he just needs to pick the choice with the highest expected payoff. Simple, right? Unfortunately, in practice the strategy of maximizing your expected return runs into a number of thorny issues.

First, most decisions don’t come with a finite set of nicely lined up choices. For our biggest decisions in life—finding a mate, choosing a career, and so on—it’s often hard to know exactly how many options are at our disposal. In addition, we often have limited information about how the various choices we can identify will actually affect our happiness. For all these reasons, real decision making usually fails to live up to economists’ lofty standards. Given the difficulty of maximizing payoffs, it’s no surprise that we make tons of bad mistakes all the time. What is surprising, is that we don’t just make random mistakes, we seem to make systematic mistakes. We don’t just experience a catastrophic cognitive meltdown when facing hard choices; we instead systematically switch on a set of simple (though mostly irrational) strategies to weigh those choices.

Psychologists have discovered two psychological biases that seem to be universal in humans: reference dependence and loss aversion. But what if humans are not the only species to use these poor decision-making strategies? Rather than investigating the biases of human subjects, Yale psychologist Laurie Santos and colleagues decided to test whether similar errors showed up in the decision making of one of our primate relatives: the capuchin monkey, whose last common ancestor with humans lived around 35 million years ago.

First they had to teach capuchin monkeys to use money—metal tokens they could exchange for food. Then, with carefully constructed experiments, they discovered that monkeys behaved just like humans: they thought about the market in terms of arbitrary reference points and responded to payoffs differently depending on whether the payoffs appeared to be gains or losses relative to those reference points. In this and other studies, monkeys seemed not to consider their choices in absolute terms—they made decisions differently when dealing with losses than when dealing with gains.

So, if our biases are more than 35 million years old, maybe we should stop trying to get rid of them and just learn to use them to our advantage. Read more.



About Computing Sciences at Berkeley Lab

The Lawrence Berkeley National Laboratory (Berkeley Lab) Computing Sciences organization provides the computing and networking resources and expertise critical to advancing the Department of Energy's research missions: developing new energy sources, improving energy efficiency, developing new materials and increasing our understanding of ourselves, our world and our universe.

ESnet, the Energy Sciences Network, provides the high-bandwidth, reliable connections that link scientists at 40 DOE research sites to each other and to experimental facilities and supercomputing centers around the country. The National Energy Research Scientific Computing Center (NERSC) powers the discoveries of 6,000 scientists at national laboratories and universities, including those at Berkeley Lab's Computational Research Division (CRD). CRD conducts research and development in mathematical modeling and simulation, algorithm design, data storage, management and analysis, computer system architecture and high-performance software implementation. NERSC and ESnet are DOE Office of Science User Facilities.

Lawrence Berkeley National Laboratory addresses the world's most urgent scientific challenges by advancing sustainable energy, protecting human health, creating new materials, and revealing the origin and fate of the universe. Founded in 1931, Berkeley Lab's scientific expertise has been recognized with 13 Nobel prizes. The University of California manages Berkeley Lab for the DOE’s Office of Science.

DOE’s Office of Science is the single largest supporter of basic research in the physical sciences in the United States, and is working to address some of the most pressing challenges of our time. For more information, please visit science.energy.gov.