A-Z Index | Phone Book | Careers

InTheLoop | 02.04.2013

The Weekly Newsletter of Berkeley Lab Computing Sciences

February 4, 2013

NERSC Helps Explain the Big Freeze

Scientists suspect that about 13,000 years ago, a catastrophic injection of freshwater into the North Atlantic “conveyor,” which transports warm tropical water northward, triggered a major cold spell, known as the Younger Dryas or Big Freeze. But until recently, nobody could fully explain how the freshwater got there. Using supercomputers at NERSC, two researchers many have finally solved this mystery. Read more.

NERSC Users’ Group Will Meet February 12–15

The annual meeting of the NERSC Users’ Group (NUG) will be February 12–15, 2013 at Berkeley Lab and NERSC’s Oakland Scientific Facility. All sessions will be remotely broadcast over the web.

On the first day of NUG 2013, NERSC will present status and plans and will discuss strategies for enhancing scientific discovery through high performance computation and storage. The second day will focus on trends in HPC and NERSC users’ experiences. The final two days are dedicated to training, for both beginners and experienced users, with an introduction to NERSC’s new Cray XC30 supercomputer, Edison.

For more information on NUG 2013 and a complete schedule, go here.

In the News: The Future of Energy: Batteries Included?

Batteries are a hugely important technology, reports The Economist. Modern life would be impossible without them. But many engineers find them disappointing and feel that they could be better still. Produce the right battery at the right price, these engineers think, and you could make the internal-combustion engine redundant and usher in a world in which free fuel, in the form of wind and solar energy, was the norm. That really would be a revolution.

It is, however, a revolution that people have been awaiting a long time. And the longer they wait, the more the doubters wonder if it will ever happen. The DOE’s Joint Centre for Energy Storage Research (JCESR), which includes Berkeley Lab collaborators, hopes to prove the doubters wrong. Their research makes use of a rapidly growing encyclopedia of substances created by the Materials Project, which computes at NERSC. Read more.

This Week’s Computing Sciences Seminars

Special EECS Seminar: Computer Chips that Communicate with Light: Building VLSI Chips with Integrated Silicon-Photonics

Tuesday, February 5, 3:30–4:30 pm, 531 Cory Hall (Wang Room), UC Berkeley
Vladimir Stojanovic, Massachusetts Institute of Technology

Chip design is radically changing. This period of change is a very exciting time in integrated circuit and system design. On one hand, cross-layer design approaches need to be invented to improve system performance despite CMOS scaling slowdown. On the other, a variety of emerging devices are lined-up to extend or potentially surpass the capabilities of CMOS technology, but require key innovations at the integration, circuits and system levels.

This talk describes how monolithic integration of photonic links can revolutionize the VLSI chip design, dramatically improving its performance and energy-efficiency. Limited scaling of both on-chip and off-chip interconnects, coupled with CMOS scaling slowdown have led to energy-efficiency and bandwidth density constraints that are emerging fast as the major performance bottlenecks in embedded and high-performance digital systems. While optical interconnects have shown promise in extensive architectural studies to date, significant challenges need to be overcome both in device and circuit design as well as the integration strategy.

We illustrate how our cross-layer approach guides the system design by connecting process, device and circuit optimizations to system-level metrics, exposing the inherent trade-offs and design sensitivities. Our experimental platforms demonstrate the technology potential at the system level and provide feedback to modeling and device design. In particular, we’ll describe the recent breakthroughs in monolithic photonic memory interface platform with fastest and most energy-efficient modulators demonstrated in a 45nm process node. Based on these design principles and technology demonstrations, we project that in the next decade tailored hybrid (electrical/optical) integrated systems will provide orders of magnitude performance improvements at the system level and revolutionize the way we build future VLSI systems.

Operating System Support for High-Throughput Processors

Tuesday, February 5, 3:30–5:00 pm, 380 Soda Hall, UC Berkeley
Mark Silberstein, UT-Austin

The processor landscape has fractured into latency-optimized CPUs, throughput-oriented GPUs, and soon, custom accelerators. Future applications will need to cohesively use a variety of hardware to achieve their performance and power goals. However, building efficient systems that use accelerators today is incredibly difficult.

In this talk we will argue that the root cause of this complexity lies in the lack of adequate operating system support for accelerators. While operating systems provide optimized resource management and Input/Output (I/O) services for CPU applications, they make no such services available to accelerator programs.

We propose GPUfs — an operating system layer which enables access to files directly from programs running on throughput-oriented accelerators, such as GPUs. GPUfs extends the constrained GPU-as-coprocessor programming model, turning GPUs into first-class computing devices with full file I/O support. It provides a POSIX-like API for GPU programs, exploits parallelism for efficiency, and optimizes for access locality by extending a CPU buffer cache into physical memories of all GPUs in a single machine.

Using real benchmarks we show that GPUfs simplifies the development of efficient applications by eliminating the GPU management complexity, and broadens the range of applications that can be accelerated by GPUs. For example, a simple self-contained GPU program which searches for a set of strings in the entire tree of Linux kernel source files completes in about third of the time of an 8-CPU-core run.

Joint work with Idit Keidar (Technion), Bryan Ford (Yale) and Emmett Witchel (UT Austin).

Communication-Avoiding Parallel Algorithms for Dense Linear Algebra and Tensor Computations: Scientific Computing and Matrix Computations Seminar

Wednesday, February 6, 12:10–1:00 pm, 380 Soda Hall, UC Berkeley
Edgar Solomonik, UC Berkeley

The motivating electronic structure calculation methods for this work are Density Functional Theory (DFT), which employs dense linear algebra, and Coupled Cluster, a method for highly correlated systems, which relies heavily on contractions of symmetric tensors. I will introduce 2.5D algorithms, an extension of 3D algorithms, which are designed to minimize communication between processors. These parallel algorithms employ limited data-replication to asymptotically lower communication costs with respect to standard (ScaLAPACK/Elemental) 2D algorithms. In particular, we can reduce the amount of data sent along the critical path of execution in matrix multiplication, LU, Cholesky, and QR factorizations, triangular solve, and the symmetric eigenvalue problem. The amount of messages sent is reduced for some of these algorithms but increased for others. This interesting discrepancy will be justified by lower-bound proofs which show the interdependence of latency and bandwidth costs. The algorithms are practical, which we demonstrate by presenting large-scale parallel results of a subset of these algorithms.

Time permitting, I will go on to discuss the extension of these method to tensor contractions and considerations for multi-dimensional symmetries. These new tensor contraction algorithms are being implemented in a new parallel software library, Cyclops Tensor Framework, which already supports Coupled Cluster with single and double excitations (CCSD). This software scales on the new BlueGene/Q architecture and outperforms the CCSD implementation of NWChem on Cray XE6.

TRUST Security Seminar: Phone Phreaks: What Can We Learn from the Earliest Network Hackers?

Thursday, February 7, 1:00–2:00 pm, 430 Soda Hall (Wozniak Lounge), UC Berkeley
Phil Lapsley

Before smartphones and iPads, before the Internet or the personal computer, a misfit group of technophiles, blind teenagers, hippies, and outlaws figured out how to hack the world’s largest machine: the telephone system. Phil Lapsley will trace the birth of the telephone, the rise of AT&T’s monopoly, the discovery of the Achilles heel in Ma Bell’s network, and the advent of the kids and outlaws — the “phone phreaks” — who hacked the telephone network for fun and profit in the 1960s and 1970s. He will then draw some connections between the phone hackers of yore and more recent network hacking incidents, such as the Aaron Swartz case.

Special EECS Seminar: How to Make Predictions When You’re Short on Information

Thursday, February 7, 3:00–4:00 pm, 430 Soda Hall (Wozniak Lounge), UC Berkeley
Benjamin Recht, University of Wisconsin-Madison

With the advent of massive social networks, exascale computing, and high-throughput biology, researchers in every scientific department now face profound challenges in analyzing, manipulating and identifying behavior from a deluge of noisy, incomplete data. In this talk, I will present a unifying optimization framework to make such data analysis tasks less sensitive to corrupted and missing data by exploiting domain specific knowledge and prior information about structure.

Specifically, I will show that when a signal or system of interest can be represented by a combination of a few simple building blocks—called atoms—it can be identified with dramatically fewer sensors and accelerated acquisition times. For example, radar signals can be decomposed into a sum of elementary propagating waves, metabolic dynamics can be analyzed as sums of multi-index data arrays, and aggregate rankings of sports teams can be written as sums of a few permutations. In each application, the challenge lies not only in defining the appropriate set of atoms, but also in estimating the most parsimonious combination of atoms that agrees with a small set of measurements.

This talk advances a framework for transforming notions of simplicity and latent low-dimensionality into convex optimization problems. My approach builds on the recent success of generalizing compressed sensing to matrix completion, creating a unified optimization framework that greatly extends the catalog of objects and structures recoverable from partial information. This framework provides a standardized methodology to sharply bound the number of observations required to robustly estimate a variety of structured models. It also enables focused algorithmic development that can be deployed in many different applications, a variety of which I will detail in this talk. I will close by demonstrating how this framework provides the abstractions necessary to scale these optimization algorithms to the massive data sets we now commonly acquire.

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.