InTheLoop | 03.09.2009
The weekly newsletter for Berkeley Lab Computing Sciences
March 9, 2009
Last Week’s ASCAC Presentations Available on the Web
The Advanced Scientific Computing Advisory Committee (ASCAC) met last Tuesday and Wednesday (March 3–4) in Washington, DC. Eight presentations from the meeting, including presentations by Juan Meza and Steve Cotter, are available here:
- View From Germantown
Michael Strayer, Associate Director, OASCR, Office of Science
- International Collaboration in High Performance Computing
Pete Beckman, Argonne National Laboratory
- Establishing Petascale Scientific Computing
Buddy Bland, Oak Ridge National Laboratory
- 2008 Gordon Bell Prize - DCA++
Thomas Schulthess, Oak Ridge National Laboratory and the Swiss National Supercomputing Center
- 2008 Gordon Bell Special Prize
Juan Meza, Lawrence Berkeley National Laboratory
- Changes to the INCITE Allocation Process
Barbara Helland, OASCR, Office of Science
- Update on the Scientific Challenges Workshop Series
Paul Messina, Argonne National laboratory
- ESnet Update
Steve Cotter, Lawrence Berkeley National Laboratory
ESnet Has Opening for Software and Network Engineer
Berkeley Lab is seeking a seasoned Software and Network Engineer to join the Energy Sciences Network (ESnet) new Advanced Network Technologies Group. This engineer will:
- Be a lead software developer on a wide range of network-related projects, including projects related to dynamic virtual circuits, network monitoring and troubleshooting, etc.
- Assist with proof-of-concept demonstrations of new network technologies.
- Assist with testing new technologies to make sure they are ready for production use.
- Interact and collaborate with researchers at LBNL, as well as from other DOE laboratories and academic institutions.
- Work with ESnet users to realize the highest possible network throughput.
NERSC Seeks High Performance Computing Consultant
The NERSC User Services Group has an opening for a High Performance Computing Consultant. This position provides technical expertise to bridge SciDAC and NERSC by engaging and supporting SciDAC science teams and enabling them to work efficiently and effectively at NERSC. Duties include:
- Develop solutions for user problems and provide one-on-one code debugging, optimization and porting help. Resolve and manage consulting questions in a timely manner and follow up on longer-term problems, working with vendors and other NERSC staff to implement solutions. Convey user feedback and concerns to other NERSC staff.
- Educate and train users by creating content for the NERSC website with online tutorials and documentation, giving presentations, and attending conferences. Communicate with users about new opportunities and capabilities in software and systems and advise them in effectively transitioning to new technologies.
- Keep abreast of developments and research in the high-performance computing field, writing technical papers as appropriate.
- Help install, maintain, document and support third-party applications, libraries and tools at NERSC, keeping such software up to date. Evaluate new software requests and assess users’ software needs.
Postdocs Will Be Invited to Take Survey
All postdocs at Berkeley Lab will soon be invited to participate in the Berkeley Lab Postdoc Survey, which begins tomorrow (Tuesday, March 10). The purpose of the survey is to gauge postdoc opinions about their supervisor, team, and work environment at Berkeley Lab and to help prioritize our efforts for postdocs by identifying what is most important to them.
The survey, which is being administered by MOR Associates, should take about 15 minutes to complete. Responses will be completely confidential (no one at Berkeley Lab will see individual responses), and all responses will be reported in aggregate by division.
See the Berkeley Lab Postdoc Web Page for more information about the survey and other programs for postdocs.
ASCR Discovery: Unlocking Thorny Ethanol Enigma
A new feature story in the online magazine ASCR Discovery, “Planting seeds to unlock thorny ethanol enigma,” describes the INCITE project of Jeremy Smith and his fellow researchers, who are trying to understand the barriers that keep biomass from becoming an economical source of ethanol. Smith, director of the Center for Molecular Biophysics, an Oak Ridge National Laboratory-University of Tennessee joint project, and his colleagues create computational models of lignocellulose. Those models could help us understand what makes this tough biomass component so difficult to break down into sugars for conversion to ethanol.
Call for SC09 Tutorial Proposals
Experts in high performance computing are invited to share their expertise with the HPC community by submitting proposals for tutorials at the SC09 conference to be held in Portland, Oregon, November 14–20, 2009. The SC09 tutorials program will give attendees the opportunity to explore a wide variety of important topics related to high performance computing, networking, and storage.
SC09 invites proposals for introductory, intermediate, and advanced tutorials, either full-day (six hours) or half-day (three hours). Submissions for tutorials and other aspects of the SC09 technical program open Monday, March 16, 2009. The deadline for submission is April 6, 2009. Go here for detailed information.
Abstract Submissions Now Open for ICAP 2009
The 2009 International Computational Accelerator Conference (ICAP) will be held August 30 to September 4, 2009 in San Francisco, organized by Berkeley Lab and SLAC. This is the primary meeting for computational accelerator physics, and a special session is being planned to focus on the computational science aspects of accelerator modeling and simulation. The deadline for abstract submissions is Friday, May 1. For more information, contact Kwok Ko.
An Evening with Bob Metcalfe at the Computer History Museum
On Tuesday, March 10, at 7:00 pm, the Computer History Museum in Mountain View will present “An Evening with 2008 Fellow Awardee Bob Metcalfe in Conversation with Cisco SVP Kathy Hill.” Metcalfe was awarded a 2008 Fellow Award by the Computer History Museum for leading the invention, standardization, and commercialization of the Ethernet local-area networking system for PCs.
Currently a partner at Polaris Ventures, Metcalfe has had three careers in technological innovation: inventing Ethernet at the Xerox Palo Alto Research Center, founding 3Com Corporation, and being a publisher-pundit as CEO of IDG’s InfoWorld Publishing Company. Metcalfe will sit down with Cisco Senior Vice President Kathy Hill to discuss his experiences in the technology industry, life lessons and current passions. He has been very active in finding a solution for the world’s energy challenge and identifying the innovations (and innovators) who will meet that challenge. Click here for more information.
This Week’s Computing Sciences Seminars
Accelerating Astrophysical Particle Simulations with Programmable Hardware (FPGA and GPU)
Tuesday, March 10, 11:00 am–noon, 50A-5132
Rainer Spurzem, ARI-ZAH, University of Heidelberg, Germany
The implementation of N-body and SPH codes on new types of accelerator hardware (field programmable gate arrays, FPGA, and graphical processing units, GPU) is presented. Our present main astrophysical applications are stellar dynamical evolution of galactic nuclei with central black holes and gravitational wave generation, and galactic dynamics with feedback and gas and feedback using SPH. The code performance on a single node using the different kinds of special hardware (traditional GRAPE, FPGA, and GPU) and some implementation aspects (e.g., accuracy) are given. We find in accord with previously published results of other groups that for N-body simulations (real application codes) GPU hardware can be used with a very high efficiency and sustained speed, for low price. FPGA is more useful for complex pipelines (like SPH), where they reach much better efficiency and slightly better sustained speed than GPU. Finally in an outlook it is argued that future faster and larger generations of FPGA may provide comparable or even higher computing power than GPU with much less power consumption — so they are a promising path in the context of the new paradigm of “green computing.” Next plans to build new types of clusters are discussed.
CHESS Seminar: Manycore Vector-Thread Architectures
Tuesday, March 10, 4:00–5:00 p.m., 540 Cory Hall, UC Berkeley
Christopher Batten, UC Berkeley
Serious technology issues are breaking down the traditional abstractions in computer engineering. Power and energy consumption are now first-order design constraints and the road map for standard CMOS technology has never been more challenging. In response to these technology issues, computer architects are turning to multicore and manycore processors where tens to hundreds of cores are integrated on a single chip. However, this breaks down the traditional sequential execution abstraction forcing software programmers to parallelize their applications. This talk will introduce a new architectural approach called vector-threading (VT) which is a first step to addressing these challenges. Vector-threading combines the energy-efficiency and simple programming model of vector execution with the flexibility of multithreaded execution.
This talk will also describe two implementations of vector-threading. The Scale VT Processor is a prototype for embedded applications implemented in a TSMC 0.18 um process. Scale includes a RISC control processor and a four-lane vector-thread unit that can execute 16 operations per cycle and supports up to 128 active threads. The 16 sq mm chip runs at 260 MHz while consuming 0.4–1.1 W across a range of kernels. We have leveraged our insights from our first implementation of vector-threading to begin developing the Maven VT Processor. A Maven chip would include tens to hundreds of simple control processors each with its own single-lane vector-thread unit (VTU). A Maven single-lane VTU is potentially easier to implement and more efficient than a Scale multiple-lane VTU. Maven lanes can be coupled together with a combination of low-level software running on the control processor and fast hardware barriers.
Par Lab Seminar: Efficient Data-Parallel Computing on Small Heterogeneous Clusters
Thursday, March 12, 11:00 am–noon, 430-438 Soda Hall, UC Berkeley
Rebecca Isaacs, Microsoft Research Cambridge
Cluster-based data-parallel frameworks such as MapReduce, Hadoop, and Dryad are increasingly popular for a large class of compute-intensive tasks. Although such systems are designed for large-scale clusters, they also offer a convenient and accessible route to data-parallel programming for small-scale clusters. This potentially allows applications traditionally targeted at supercomputers or remote server farms, such as sophisticated video processing, to be deployed in a small-scale ad-hoc fashion by aggregating the servers and workstations in the home or office network.
The default scheduling algorithms of these frameworks perform well at scale, but are significantly less optimal in a small (3-10 machine) cluster environment where nodes have widely differing performance characteristics. To make effective use of an ad-hoc cluster, we require a “planner” rather than a scheduler that takes account of the predicted resource consumption by each vertex in the dataflow graph and the heterogeneity of the available hardware.
In this talk I will describe our enhancements to DryadLINQ and Dryad for ad-hoc clusters. We have integrated a constraint-based planner that maps the dataflow graph generated by the DryadLINQ compiler onto the cluster. The planner makes use of DryadLINQ operator performance models that are constructed from low-level traces of vertex executions. The performance models abstract the behaviour of each vertex in sufficient detail to predict the bottleneck resource, which can change during vertex execution, on different hardware and with different sizes of input. Experimental evaluation shows reasonable predictive accuracy and good performance gains for parallel jobs on ad-hoc clusters.
NERSC Lunch Talk: IO Performance Testing for a Climate Modeling Application
Thursday, March 12, noon–1:30 pm, OSF Room 238
Mark Howison, LBNL/CRD and UC Berkeley
The Visualization Group at LBNL has been providing ongoing support with IO and visualization strategies for Dave Randall’s INCITE19 climate modeling project. Working with researchers at PNNL who are implementing a library for outputting geodesic grid data, we have uncovered several bottlenecks that lead to suboptimal parallel IO performance on Franklin, much below the system’s theoretical peak. I will give an overview of the application, the layout of the data, and the visualization techniques, followed by an in-depth analysis of IO performance issues we have identified using tools such as IOR and IPM.
Green Flash: Power Efficient Scientific Computing at the Exascale
Thursday, March 12, 4:00–5:00 pm, 290 Hearst Memorial Mining Bldg., UC Berkeley
Michael Wehner, LBNL/CRD
Since the first numerical weather experiments in the late 1940s by John Von Neumann, machine limitations have dictated to scientists how nature could be studied by computer simulation. In the intervening years, scientific computing has grown into a mature and sophisticated endeavor with many fundamental discoveries to its credit in a wide variety of disciplines. However, the remarkably steady increase in scientific computing power since Von Neumann’s time is now threatened. As the processors used in traditional massively parallel computers have become ever more complex, electrical power demands are approaching hard limits dictated by cooling considerations. It is clear from both the cooling demands and the electricity costs, the growth in scientific computing capabilities of the last few decades is not sustainable unless fundamentally new ideas are brought to bear.
In this talk, we propose a radically new approach to high-performance computing (HPC) design via application-driven hardware and software co-design which leverages design principles from the consumer electronics marketplace. As part of our approach, architecture and algorithm design are inextricably entwined. We find, via the example scientific problem of constructing a cloud system resolving climate model, that exaflop class scientific computing can be done at far lower capital and operational costs than traditional methods.
Link of the Week: Thirsty Work
Water can be a tough subject to firmly grasp. For example, state water officials continue to talk of drought, while many Bay Area residents have had enough rain for the time being, thank you. In a recent issue, The Economist took a different view of water — examining how much H2O is required to make some of our favorite products. Taking two products near and dear to Computing Sciences staff, for example, which takes more water to produce: a kilo of coffee or a kilo of microchips? A liter of brewed coffee or a liter of brewed beer? Or a kilo of potato chips or a kilo of microchips? Find the answers and more here.
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.