InTheLoop | 03.30.2015
Simulations Predict Arctic Warming = Net CO2 Increase
There’s a carbon showdown brewing in the Arctic as Earth’s climate changes. On one side, thawing permafrost could release enormous amounts of long-frozen carbon into the atmosphere. On the opposing side, as high-latitude regions warm, plants will grow more quickly, which means they’ll take in more carbon from the atmosphere.
According to new computer simulations conducted at NERSC by Berkeley Lab scientist, there will be a lot more carbon released from thawing permafrost than the amount taken in by more Arctic vegetation. This net release of carbon dioxide into the atmosphere could accelerate climate change. »Read more.
GCN Article Features NERSC's Hick on Storage Strategy
Government Computing News recently featured NERSC Group Lead for Storage Systems Jason Hick in an article tackling best practices for massive storage systems. In the article, Hick is quoted extensively on strategies for balancing storage technologies for the best mix of economy and responsiveness:
“There are a variety of emerging storage technologies. Using all of them for what they do best is key. You need to understand flash, disk and tape and come up with the best mix based on your internal operations. It pays huge dividends for us when we balance our workloads well.” »Read more.
Reminder: Friday Deadline for ATPESC 2015
Doctoral students, postdocs, and early career computational scientists and engineers are encouraged to apply to the Argonne Training Program on Extreme-Scale Computing (ATPESC). Held August 2-14, 2015, this intensive two-week course is designed to train the next generation of supercomputer users. The deadline to apply is Friday, April 3. The program is free and domestic airfare, meals and lodging are provided for the 60 accepted participants. »Learn more and apply.
This Week's CS Seminars
Highly scalable asynchronous algorithm for partial differential equations: a path towards Exascale
Monday, March 30, 1:30 - 2:30pm, Bldg. 50F, Room 1647
Aditya Konduri, Department of Aerospace Engineering, Texas A&M University
Synchronization overheads pose a major challenge as applications advance towards extreme scales. In current algorithms for partial differential equations (PDEs), synchronization as well as data communication delay the parallel computations at each time step and may create a new scaling wall when moving towards Exascale. In this work, we present an asynchronous computing algorithm based on finite difference schemes for PDEs where no synchronization between processing elements (PEs) is enforced. PEs are allowed to continue computations regardless of message status and are thus asynchronous. We show that accuracy of commonly used finite difference schemes is degraded when they are used asynchronously. Since message arrival at PEs is essentially a random process, so is the behavior of the error. Within a statistical framework we show that average errors drop always to first-order regardless of the original scheme. The value of the error is found to depend on both grid spacing as well as characteristics of the computing system including number of PEs and statistics of the delays. We propose new schemes that are robust to asynchrony and present a novel asynchronous algorithm for their implementation. We utilize modern remote memory access programming schemes that have been shown to provide significant speedup on modern supercomputers, to efficiently implement communications suitable for asynchronous schemes. We present results from simulations that demonstrate excellent scalability of the new asynchronous schemes for large-scale computing.
Remote Access via Webex
Event number: 669 443 344
Applied Mathematics-Statistical Data Assimilation: Path Integrals and Approximation
Wednesday, April 1, 2:30 – 3:30pm, 939 Evans Hall - UC Berkeley Campus
Henry Abarbanel University of California, San Diego
Formulating statistical data assimilation with noisy measurements and model errors as a path integral over the path of a model state during a temporal observation window holds advantages over existing variational principles in data assimilation: two of these are in a framework where corrections to the variational methods can be evaluated and in an approach to identifying the minimum of the variational objective function. Examples from nonlinear models using in geophysics and neurobiology will be given as well an example from analysis of laboratory experiments on neurons in the songbird vocalization system.
The Hacker Within: R-a high-level programming language for statistical analysis
Wednesday, April 1, 4 – 6pm, 190 Doe Library - UC Berkeley Campus
Rochelle Terman, Daniel Turek, Chris Paciorek
This is a weekly meeting for sharing skills and best practices for scientific computation sponsored by the Berkeley Institute for Data Science (BIDS). Based on The Hacker Within Scientific Computing Group from the University of Wisconsin–Madison, the UC Berkeley chapter uses this as a structured set of skill-sharing sessions for scientific software development (e.g., testing, data management, version control, literate programming, etc. ) The goal is to learn cool skills and incorporate these practices into our workflows. People from all scientific disciplines are welcome. This meeting would be a great venue for describing neat tips and tricks for efficiency, introducing new libraries, showing off useful features of a scientific code you're using, or bringing up a computational problem you're having.
This week's topic is R, a high-level programming language for statistical analysis. »Learn more.
Cosmology and formation of structure in the universe
Thursday, April 2, 12 - 1pm, Bldg. 50F, Room 1647
Zarija Lukic, Computational Cosmology Center, Berkeley Lab
The speaker will give a broad overview of the field: What have we learned about the universe so far? What are some things we would like to know? Why does it matter? What roles do high-performance computing, applied mathematics and computer science play in advancing our understanding of how the universe's structure formed and in uncovering the nature of dark matter and dark energy?
BIDS Data Science Lecture: Replication, Communication, and the Population Dynamics of Scientific Discovery
Friday, April 3, 1 - 2:30pm, 190 Doe Library - UC Berkeley Campus
Paul Smaldino, Postdoctoral Scholar, UC Davis
Many published research results are false, and controversy continues over the roles of replication and publication policy in improving the reliability of research. I will present a mathematical model of scientific discovery in the context of replication, publication bias, and variation in research quality. This model provides a formal framework for reasoning about the normative structure of science. It is only a start, but it speaks directly to ongoing debates about the design and conduct of science.