InTheLoop | 07.27.2009
The weekly newsletter for Berkeley Lab Computing Sciences
July 27, 2009
DOE SC Early Career Research Program Offers Opportunities
Interim Laboratory Director Paul Alivisatos will hold an open discussion of the Office of Science Early Career Research Program on Tuesday, July 28. All interested applicants are welcome to attend the meeting in the 66 Auditorium from 3:00 to 4:00 pm. For those who cannot attend, the meeting will be recorded and made available on the Lab’s website Office of Science Early Career Research Program.
This is the first year of an annual program to support the development of individual research programs of outstanding scientists early in their careers and to stimulate research careers in the areas supported by the DOE Office of Science. Up to $60M in Recovery Act funds will be available in FY2010 in program areas including Advanced Scientific Computing Research.
Eligible PIs must be full-time, career or career track/term, non-postdoctoral national laboratory employees, no more than ten years from PhD. Letters of Intent must be submitted by August 3, 2009. Detailed information, including the DOE national laboratory announcement, FAQ, and the LBNL internal submission process can be found on the LBNL Early Career Program website.
Heeding Cyber Warning Uncovers Hacked Site on Campus
Computing Sciences conference planner Yeen Mankin was recently looking at several event venues when a visit to the Cal Faculty Club website triggered an alert on her Mac notebook. She reported her suspicions to Jay Krous of LBNL’s Computer Protection Program. Krous did a little looking around and found that the site had indeed been hacked and malware known as microsotf.cn embedded. He noted that microsotf.cn is a known malicious domain and added, “Excellent find, Yeen.”
Openings for Senior Software Developer, HPC Technician
The Applied Numerical Algorithms Group in CRD’s HPC Research Department has an opening for a Senior Software Developer. This person will lead small teams distributed across multiple organizations and sites in the design and implementation of software for structured-grid numerical methods for partial differential equations as part of the Chombo development effort. Go here for details.
NERSC’s Computer Operations and ESnet Support Group has an opening on the first shift (7:00 am to 3:00 pm) for a High Performance Computing Technician. This person will perform first line support for NERSC and ESnet including monitoring supercomputers, the mass storage system, general servers, network routers and switches, and user support. Go here for details.
The Lab’s Employee Referral Incentive Program (ERIP) awards $1,000 (net) to employees whose referral of an external candidate leads to a successful hire.
Time Changes for CS All-Hands Meetings on August 5
Computing Sciences will be holding two all-hands meetings and a pizza party on Wednesday, August 5. You can attend either of the two meetings: from 11:30 to 12:30 or from 1:30 to 2:30 in the 70A-3377 conference room. (Note that the starting times have changed since last week’s announcement.) The pizza party on the cafeteria lawn is scheduled from 12:30 to 1:30.
This Week’s Computing Sciences Seminars
Improving Throughput of Simultaneous Multithreaded (SMT) Processors using Shareable Resource Signatures and Hardware Thread Priorities
Tuesday, July 28, 9:30–10:30 am, 943-236 and 50B-2222
Mitesh Meswani, University of Texas, El Paso
Simultaneous multithreading (SMT) allows multiple applications to execute concurrently on a processor core, potentially increasing the utilization and throughput of the processor by a factor of the degree of multithreading. However, such performance gains may not be achieved due to contention for resources shared by the applications executing in SMT mode. The IBM POWER5, which has two hardware threads per core, provides software-controlled hardware thread priorities that facilitate the control of the ratio of decode cycles allocated to the two hardware threads of a core and, therefore, the degree of resource contention between the threads. By default, the priorities of the two hardware threads are set to equal values — each gets half of the decode cycles. Several studies have shown that many scientific applications do not achieve their best throughput at equal priorities. The best priority settings or best priority pair, i.e., the priority settings for a given pair of application threads that give the best throughput, depend on the characteristics of the co-scheduled application threads.
In this talk I present a methodology for predicting the best priority pair for a given co-schedule of serial applications. Our approach exploits resource-utilization information, called as Shareable Resource Signatures (signatures), collected during an application’s single-threaded execution. The utilization information provides insights about the availability of resources that are shared in SMT mode and, thus, available for the co-scheduled application’s use. Given a co-schedule of two applications and their signatures, the methodology is used to predict the best priority pair for the co-schedule. I will present the results of an implementation of my methodology on the IBM POWER5 with NAS NPB3.2 serial benchmarks, SPEC CPU2006, and PETSc KSP solver library.
Behavior-Based Program Analysis and Parallelization
Wednesday, July 29, 11:00 am–12:00 pm, 50F-1647
Chen Ding, Associate Professor, Department of Computer Science, University of Rochester
Large-scale memory usage and high-level parallelism are difficult to analyze with traditional program-based analysis because these are aggregate phenomena depending on not only program code but also program input, the execution environment, and the use of external libraries. A growing body of research is taking a behavior-based approach, which is to analyze sample executions of a program and build a statistical model to predict how the behavior changes in other executions. Conceptually, this training-based analysis is analogous to observation and prediction in the physical and biological sciences. Through this approach, various studies in recent years have pieced together a systematic view of program locality in sequential code. In this talk I will first present a survey of these findings, including a few useful tools for memory performance tuning.
Recent work in my group explores suggestion-based programming for improving program behavior. Suggestions express likely rather than definite behavior and may not always be correct or complete. To ensure correctness, our systems use additional processors on modern machines to monitor program execution and preserve the original semantics. As an example, I will describe the design and implementation of a minimalistic parallel language, which we have used to parallelize a set of large legacy C applications including a use of a commercial math library.
The presentation is based on material from two upcoming publications.
Link of the Week: Chimpanzees, Bonobos, and Humans
In the six million years since hominids split from the evolutionary ancestor we share with chimpanzees and bonobos, something happened to our brains that allowed us to become master cooperators, accumulate knowledge at a rapid rate, and manipulate tools to colonize almost every corner of the planet. In evolutionary time, our progress has been swift and ruthless. What allowed us to come down from the trees, and why? Vanessa Woods and Brian Hare address these questions in “Out of Our Minds: How Did Humans Come Down from the Trees and Why Did No One Follow?”
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.