A-Z Index | Phone Book | Careers

InTheLoop | 02.09.2015

February 9, 2015

New Scientific Software Catalog Now Available

Researchers in Berkeley Lab’s Computational Research Division (CRD) are renowned for developing and contributing to novel software packages for use in modeling and simulation, computer science and data science.  For the first time, all of these tools have been incorporated into a single catalog on the CRD website where users can easily search and download what they need. “This catalog represents decades of Berkeley Lab expertise and leadership in applied math, computational science, computer science, data science and technology,” said CRD Director David Brown. »Read more.

Annual NERSC Users Group Meeting to Feature User Training, Science Presentations

The annual NERSS Users Group (NUG) meeting will be held Feb. 23-26 at the Oakland Scientific Facility (OSF) and the main Berkeley Lab campus.

Monday, Feb. 23 will be a training session for new users and an overview of data and analytics topics, and Wednesday, Feb. 25 will feature a kernel-tuning hackathon. The NUG Executive Committee will hold a business meeting on Thursday, Feb. 26. All of these sessions will be held at OSF in downtown Oakland.

On Tuesday, Feb. 24, there will be series of science talks in the Bldg. 50 auditorium at Berkeley Lab. Among the presentations are keynote talks on “Efficient modeling of laser-plasma accelerators using the ponderomotive-based code INF&RNO” by Carlo Benedetti of Berkeley Lab’s BELLA Center and the LOASIS Program, and “Transforming Beamline Science with SPOT Suite” by CRD’s Craig Tull. NERSC’s Katie Antypas will discuss “Cori: A Next-Generation System for Computational and Data Science” and CRD’s Phil Colella will give a talk on “Programming Next-Generation HPC Systems.” »See the full meeting agenda.

Microsoft Azure Training for Researchers

Microsoft Research and Berkeley Institute for Data Science (BIDS) at UC Berkeley are hosting a no-cost, hands-on technical training about Windows Azure on Wednesday, February 11. The one-day course will take place from 9 a.m. – 5 p.m. at the BIDS workspace: 190 Doe Library on the UC Berkeley campus. All attendees must register in advance to secure a place.

As a participant, you will be able to access Windows Azure on your own laptop during the training and will receive a three-month Windows Azure pass you can experiment with and develop your work using the cloud. You will also learn more about receiving significant Windows Azure 12-month grants for your research.

The training is part of a partnership between Microsoft Research and BIDS to educate and support the global research community in the use of cloud computing in science. »Learn more.

Former Alvarez Fellow Didem Unat Receives Prestigious European Grant

Didem Unat, a computer scientist who was an Alvarez Fellow at Berkeley Lab from 2012-14, has been awarded a Marie Skłodowska-Curie Career Reintegration grant from the European Commission. The grant covers two years of research costs for researchers who are becoming professionally reintegrated at an institution typically in their country of origin. In September 2014, Unat was appointed as a professor in the Computer Science and Engineering Department at Koç University in Istanbul in her native Turkey. In the past, the acceptance rate for the grant was just three percent.

This Week's CS Seminars

»CS Seminars Calendar

Understanding photon spectroscopies in correlated materials by developing massively parallel numerical methods

Monday, Feb. 9, 1–2 p.m., Bldg. 50F, Room 1647
Chunjing Jia, Department of Applied Physics, Stanford University

Correlated materials exhibit emergent electronic and magnetic properties due to the localization of electrons in partially filled d or f orbitals. This makes the materials hard to understand due to the enhanced Coulomb interaction. Recently, the rapid development of synchrotron light sources has enabled photon spectroscopies to become powerful tools to study these materials as the dynamics can be probed directly. These techniques provide a leap forward in experimental approaches to understanding correlated materials, but at the same time provoke challenges to theory and simulation.

In this talk, I will describe how we implement sophisticated numerical techniques to close the gap between theory and experiment for photon spectroscopies. We have developed efficient algorithms to simulate the spectroscopic cross-sections by directly accessing the many-body quantum mechanical wavefunctions. I will highlight the use of these massively parallel numerical algorithms to understand resonant inelastic x-ray scattering and x-ray absorption spectroscopy in transition metal oxides.

hStreams: A Streaming library for heterogeneous platforms

Tuesday, Feb. 10, 10:30 a.m.–12 p.m., Bldg. 50F, Room 1647
Chris (CJ) Newburn, Intel, Inc.

Making parallel programming easier, for platforms of increasing complexity and diversity, is one of the greatest challenges we face in high-performance computing.  Simplifying the management of dependences among tasks is a first step. We try out one of the simplest abstractions: a FIFO-like stream, and see how far we can carry that. We also abstract physical platform properties at a logical layer that the programmer interacts with. This abstraction greatly eases portability, but still enables total control, if users desire that.

This talk describes hStreams, a library that enables concurrency across and within nodes in heterogeneous systems, and between communication and computation.  hStreams is released as part of the Intel® Many-integrate core Platform Software Stack, and is starting to be used by academics and independent SW vendors.  We invite a lively discussion on problems and opportunities to which hStreams may be applied and alternative solutions.

Algorithms for stencil computation

Wednesday, Feb. 11, 2–3 p.m., Bldg. 50F, Room 1647
Yuan Tang, Fudan University, Shanghai, China

A stencil computation repeatedly updates each point of a $d$-dimensional grid as a function of itself and its near neighbours. In this talk, we will first talk about "Pochoir," a domain-specific compiler and run-time system that allows a programmer to write a simple specification of a stencil in a domain-specific language embedded in C++ which the Pochoir compiler then translates into high-performing Cilk code that employs an efficient cache-oblivious algorithm. Pochoir supports general \(d\)-dimensional stencils and handles both periodic and aperiodic boundary conditions in one unified algorithm. The algorithm behind Pochoir was based on the notion of "trapezoidal decomposition", but improves on prior cache-efficient algorithms on multi-dimensional grids by making "hyper-space" cuts, which yield asymptotically more parallelism for the same cache efficiency. Our recent research reveals that by scheduling subtasks derived from recursive divide-and-conquer ready to execute as soon as all its dependencies are satisfied can further improve the span (depth) of stencil computations on a \(d\)-dimensional hypercubic grid of width \(w\) for \(h\) time steps from $\Theta((d^2 h) w^{\log (d+2) - 1})$ to $\Theta(h)$. Combining a fine-grain scheduling policy with a recursive divide-and-conquer approach, we achieve best cache complexity and optimum span at the same time.

Applied Mathematics Seminar: A variational perspective on wrinkling patterns in thin elastic sheets

Thursday, Feb. 12, 4:10 p.m. – 5 p.m., 60 Evans Hall, UC Berkeley
Robert Kohn, Courant Institute of Mathematical Sciences

Thin sheets exhibit a daunting array of patterns. A key difficulty in their analysis is that while we have many examples, we have no classification of the possible "patterns." I have explored an alternative viewpoint in a series of recent projects with Peter Bella, Hoai-Minh Nguyen, and others. Our goal is to identify the scaling law of the minimum elastic energy (with respect to the sheet thickness, and the other parameters of the problem). Success requires proving upper bounds and lower bounds that scale the same way. The upper bounds are usually easier, since nature gives us a hint. The lower bounds are more subtle, since they must be ansatz-independent. In many cases, the arguments used to prove the lower bounds help explain "why" we see particular patterns. My talk will give an overview of this activity, and details of some examples.

The Birth of Boolean Logic

Even if you've never heard of mathematician George Boole, you are still beholden to him for the logic that underlies search engine queries, and for that matter, most programming languages. The self-taught son of a tradesman, Boole became a professor at Queen's College (now the University College Cork) in 1849 at age 33, with no degree of his own. While he made breakthroughs in several areas of mathematics, it is Boolean logic (or symbolic logic) for which he’s best known today. It combines logic and algebra to reduce all logical problems to symbols, including the operators most of us have learned to use in the advanced options of search engines (AND, OR, and AND NOT). It is in computer programming that Boole's logic has proven most useful — or at least most used. This year University College Cork is celebrating the 200th anniversary of Boole's birth with a site that explores all facets of his legacy, from mathematics to computer programming, engineering, philosophy and even art and literature. »Visit "Boole 200" to learn more.