InTheLoop | 04.27.2015
NERSC, Cray Move Forward With Next-Generation Scientific Computing
NERSC and Cray Inc. announced last week that they have finalized a new contract for a Cray XC40 supercomputer that will be the first NERSC system installed in the newly built Computational Research and Theory facility at Lawrence Berkeley National Laboratory.
This supercomputer will be used as Phase 1 of NERSC’s next-generation system named “Cori” in honor of bio-chemist and Nobel Laureate Gerty Cori. Expected to be delivered this summer, the Cray XC40 supercomputer will feature the Intel Haswell processor. The second phase, the previously announced Cori system, will be delivered in mid-2016 and will feature the next-generation Intel Xeon Phi™ processor “Knights Landing,” a self-hosted, manycore processor with on-package high bandwidth memory that offers more than 3 teraflop/s of double-precision peak performance per single socket node.
“This is an exciting year for NERSC and for NERSC users,” said Sudip Dosanjh, director of NERSC. “We are unveiling a brand new, state-of-the-art computing center and our next-generation supercomputer, designed to help our users begin the transition to exascale computing. Cori will allow our users to take their science to a level beyond what our current systems can do.” »Read more.
CRD's Almgren Takes the Stage for 'Science at the Theater' Event
Every discovery and invention starts with a question from simulating dark matter and cleaner burning fuels, to growing food in our cities, to the enormity of tackling the "emperor of all maladies," cancer. At 7 p.m. on Wednesday, April 29, Ann Almgren will join four other Berkeley Lab scientists onstage at Oakland's Kaiser Center Auditorium where they will dive into the big questions driving their research.
Almgren, who is the acting lead for the Computational Research Division's Center for Computational Sciences and Engineering, or CCSE, develops algorithms and software used to delve into mysteries large (the invisible scaffolding of our universe) and small (the dynamics at the heart of a flame). Joining her will be lab researchers Judy Campisi, speaking on cancer and aging; Kai Vetter, addressing radiation and public safety; Shashi Buluswar, explaining the Urban Food Intiative; and Javier Ceja-Navarro speaking on "beetles, biofuels and coffee."
Sponsored by the Friends of Berkeley Lab, the event is free (»RSVP online) and the public is encouraged to attend. »See videos of past events.
Lab Staff Present Coupled Science Facility Model at Internet2 Conference
Lab staff from five divisions will share their expertise in a panel discussion on “Creating Super-facilities: a Coupled Facility Model for Data-Intensive Science” at the Internet2 Global Summit to be held April 26-30 in Washington, D.C. The panel was organized by Lauren Rotman of ESnet and includes Alexander Hexemer of the Advanced Light Source (ALS), Craig Tull of CRD, David Skinner of NERSC and Rune Stromsness of the IT Division.
The session highlights the concept of a coupled science facility or “super-facility," a new model that links together experimental facilities like the ALS with computing facilities like NERSC via a Science DMZ architecture and advanced workflow and analysis software, such as SPOT Suite developed by Tull’s group. Session participants will share best practices, lessons learned and future plans to expand this effort.
Also at the conference, ESnet’s Brian Tierney will participate in a session on “perfSONAR: Meeting the Community's Needs.” Co-developed by ESnet, perfSONAR is a tool for end-to-end monitoring and troubleshooting multi-domain network performance. The session includes an overview and update on the perfSONAR project, including an overview of the 3.4 release, a preview of the 3.5 release, an overview of the product plan, and an overview of perfSONAR training plan.
ESnet’s Zurawski, Buraglio to Give Cybersecurity Webinars to Great Plains Regional Networks
Jason Zurawski of ESnet’s Science Engagement team will present a May 8 webinar on cybersecurity to members of the Great Plains Network, a consortium of universities in the Great Plains region. Zurawski’s talk will include a basic overview of cyber threats to a campus, and ways the Science DMZ architecture can be implemented to protect and allow high-performance activities.
On May 22, Nick Buraglio of Esnet’s Network Engineering Group will give a webinar on the basics of Bro, the intrusion detection system first developed at Berkeley Lab.
The talks are part of a series of webinars ESnet staff are presenting to members of the consortium.
ESnet Co-organizing Midwest Workshop on Innovative Network Operations
ESnet experts, along with Indiana University and Internet2 staff, are presenting a two-day workshop to members of the Great Plains Network consortium on May 27-28, in Kansas City, Mo. The workshop is being held in conjunction with the annual meeting of the Great Plains Network, a consortium of universities in the Great Plains region. The workshop will focus on Science DMZ network architectures, perfSONAR performance measurement software, Data Transfer Nodes, and emerging Software Defined Networking techniques. Combined, these technologies are proven to support high-performance, big data science applications, including high-volume bulk data transfer, remote experiment control and data visualization.
This Week's CS Seminars
Jump splicing schemes for elliptic interface problems and the incompressible Navier-Stokes equations
Monday, April 27, 2:00pm - 3:00pm , Bldg. 50F, Room 1647
Ben Preskill, UC Berkeley and Berkeley Lab
We present a general framework for accurately evaluating finite difference operators in the presence of known discontinuities across an interface. We use these ideas to develop simple to implement, second-order accurate methods for the Poisson equation with interfacial discontinuities and the incompressible Navier-Stokes equations with singular forces. We represent the interface by a level set function and locally modify the discontinuous quantities using dimension- and coordinate-independent normal Taylor expansions of the jump conditions. We then arrive at a versatile expression relating the derivatives being evaluated, the finite difference stencil, and the extrapolation. Our method is tolerant to non-smooth geometry, permits the use of symmetric positive-definite solvers for elliptic equations, and works in both 2D and 3D. We rigorously establish the convergence properties of the method and present extensive numerical convergence studies.
A perturbation-method-based post-processing of planewave approximations for DFT Kohn-Sham models
Wednesday, April 29, 2:30pm - 3:30pm, 939 Evans Hall - UC Berkeley Campus
Benjamin Stamm, Laboratoire Jacques-Louis Lions, Paris
In this talk we consider a post-processing of planewave approximations for DFT Kohn-Sham models by considering the exact solution as a perturbation of the discrete solution. Applying then Kato’s perturbation theory leads to computable corrections with a provable increase of the convergence rate in the asymptotic range. I first focus on the key-features of this post-processing by carefully analyzing the Gross-Pitaevskii equation that serves as a toy problem before I discuss to the DFT Kohn-Sham models. Finally some numerical illustrations are presented. If time permitting, I will also discuss some recent advances of a posteriori estimates for the Gross-Pitaevskii equation.
RISC-V: A Free, Open, Extensible ISA for the Heterogenous Future
Thursday, April 30, 2015, 11:00am - 12:00pm, Bldg. 50F, Room 1647
Yunsup Lee, Computer Science Department, UC Berkeley
RISC-V (pronounced "risk-five") is a new, open, and completely free general-purpose instruction set architecture (ISA) developed at UC Berkeley. RISC-V was originally designed to support computer architecture research and education, but we hope it will also become a standard open architecture for industrial implementations. The ISA is designed in a modular fashion with a base set of integer instructions plus standard extensions (e.g., multiplication/division, atomics, single-, and double-precision floating-point). In addition, there is room set aside for custom extensions to the ISA that any company could add as their “secret sauce.” The ecosystem is already significant and include ports of GCC, GDB, glibc, Newlib, Linux, and LLVM. It also offers software simulators, including an in-house ISA simulator and a QEMU port. We have fabricated multiple test chips in 28nm and 45nm process technologies that run at 1-1.65 GHz and boot Linux. The scalar core implemented in the test chips achieves a 10% higher DMIPS/MHz score than an ARM Cortex-A5, while being 49% more area-efficient. A custom vector accelerator integrated alongside the scalar core demonstrates the extensibility of the RISC-V ISA. The vector accelerator is 1.8X more energy-efficient while running double-precision matrix multiplication than the IBM Blue Gene/Q processor, and 2.6X more than the IBM Cell processor, both fabricated in the same process. The first RISC-V cores have already shipped in commercial parts, and commercial RISC-V microcontrollers will appear in silicon later this year. External projects at IIT Madras, lowRISC, and EDA company Bluespec have also recognized the potential of RISC-V, and have begun using RISC-V. For more information, please visit http://riscv.org.
Improving Performance of Structured-memory, Data-Intensive Applications on Multi-core Platforms via a Space-Filling Curve Memory Layout
Thursday, April 30 12:00pm-1:00pm, Bldg. 50F, Room 1647
Wes Bethel, DAV group, Berkeley Lab
Many data-intensive algorithms—particularly in visualization, image processing, and data analysis—operate on structured data, that is, data organized in multidimensional arrays. While many of these algorithms are quite numerically intensive, by and large, their performance is limited by memory access. Targeting the challenges associated with minimizing data movement through the memory hierarchy within a node in a shared-memory parallel setting, which is one of the research challenges facing the community as we move towards the exascale regime of computing, we study the effects that an alternative in-memory data layout format has in terms of runtime performance and utilization of the memory hierarchy. We focus the study on shared-memory parallel implementations of two algorithms common in visualization and analysis: a stencil-based convolution kernel, which uses a structured memory access pattern, and raycasting volume rendering which uses a semi-structured memory access pattern. The question we study is to better understand to what degree an alternative memory layout, when used by these key algorithms, will result in improved runtime performance and memory system utilization. Our approach uses a layout based on a Z-order (Morton-order) space-filling curve data organization, and we measure and report runtime and various metrics and counters associated with memory system utilization. Our results show nearly uniform improved runtime performance and improved utilization of the memory hierarchy across varying levels of concurrency the applications we tested. This approach is complementary to other memory optimization strategies like cache blocking, but may also be more general and widely applicable to a diverse set of applications..
BIDS Data Science Lecture Series: Computational Imaging with Nonlinear Inverse Problems
Friday, May 1, 1:00pm to 2:30pm, 190 Doe Library UC Berkeley Campus
Laura Waller, UC Berkeley
Computational imaging involves the joint design of optical systems and post-processing algorithms such that computation replaces optical elements, enabling simple experimental setups. This talk will describe new optical microscopes that employ simple experimental architectures and efficient nonlinear inverse algorithms to achieve high-resolution 3D and phase images. By leveraging recent advances in computational illumination, we achieve brightfield, darkfield, and phase contrast images simultaneously, with extension to 3D and gigapixel phase imaging. We discuss unique challenges for large-scale real-time imaging of biological samples in vitro and in vivo. Sponsored by the Berkeley Institute for Data Science.