Powell Booth 100
Monday June 2
Speaker: Mark Adams
“A Proposal for a New Top500 Metric”
The High Performance Linpack (HPL) benchmark and associated Top-500 List has
been a successful metric for ranking high performance computing systems. HPL
became a broadly accepted representative for application performance when it
came to prominence in the 1990s, but over time has become less reflective of
the system performance metrics that matter to contemporary science and
engineering applications as lower complexity algorithms have been developed
as required for extreme scale computing. We propose a replacement for HPL
that maintains many of its desirable qualities: a direct, non-iterative,
solver for systems of linear algebraic equations. We define a high
performance geometric multigrid (HPGMG) benchmark that provides a more
balanced exercise of machine capabilities and a more accurate proxy to
modern application requirements. Using HPGMG, we strive to create a
benchmark for ranking systems that will promote system design improvements
that are better aligned to real scientific application performance.
Mark Adams received his Ph.D. in Civil Engineering, from U.C. Berkeley in
1998 and was a student and postdoc with Jim Demmel in Computer Science, at
U.C. Berkeley. He is a member of the Applied Numerical Algorithms Group at
Lawrence Berkeley National Laboratory, and is an adjunct research scientist
in the Applied Physics and Applied Mathematics Department at Columbia
Powell Booth 100
CACR is proud to announce that three of our employees will be receiving recognition at Caltech’s 59th annual Service Awards Ceremony, to be held in Beckman Auditorium on Monday, June 2, 2014.
Sharon Brunett – 25 years
Mark Bartelt – 15 years
Santiago Lombeyda – 15 years
Many thanks to Sharon, Mark & Santiago for their service to the institute and to CACR!
CACR is pleased to announce the addition of another 256 compute cores to the zwicky cluster, used exclusively by Caltech’s Theoretical Astrophysics (TAPIR) group for simulation of black holes and extreme spacetimes.
During the week of January 6, 2014, 16 new Hewlett-Packard SL250 nodes, each with two Intel E5-2670 8 core processors were integrated into the existing cluster.
Zwicky’s compute configuration now has 2,564 cores – 187 dual processor Intel x5650 6 core nodes plus 20 dual processor Intel E5-2670 8 core nodes, all connected with [F,Q]DR InfiniBand.
We would like to thank the TAPIR group for continuing to fund, expand, and enhance zwicky resources.
Please see this page for an overview of CACR managed sxs compute and storage resources.
The CHOPIN project plan consists of several major elements: Deployment of SDN (Software-Defined Networking) capable switches supporting direct connectivity to the California OpenFlow Testbed Network (COTN) for Research and Development activities in the networking area, as well as the connection to the Internet2 Advanced Network Services, meant explicitly to support high-throughput data applications and research using large data volumes. The connection will also enable direct access to the nation-wide GENI infrastructure. In a direct way, the CHOPIN network will also connect to the DYNES cyber instrument, and extend its use within Caltech.
The project plan also includes the deployment of a new 100Gbps uplink between Caltech and CENIC. in support of current and future data-centric research and high-throughput applications. The new 100G capability will also pave the way for integration with routed high-performance services such as the CENIC HPR-L3 and Internet2 Advanced Layer3 Services. The uplink is complemented by 100G and 40G links to a set of Openflow capable switches to each of several major research groups.
The upgrades in capacity and capability will ensure that the Caltech campus infrastructure is up-to-date for the next few years, through direct access to the new high-capacity national backbones of Internet2, Energy Sciences Network, and possibly National LambdaRail.
The CHOPIN project infrastructure will provide fertile ground for Caltech faculty and their students as both a research tool as well as a tool for computer and computational scientists, and will benefit the entire Caltech campus, and explicitly groups involved in Big Data, high-throughput research.
CACR is pleased to announce that an award of $233,000 from the National Science Foundation Astronomy & Astrophysics Research Grants Program has been made to the Catalina Real-Time Transient Survey-II (CRTS-II). The project, entitled “Open Exploration of the Time Domain with the Catalina Real-Time Transient Survey”, will analyze data from the upgraded telescopes of Catalina Sky Survey (CSS, http://www.lpl.arizona.edu/css/), which scour the sky for potentially life-threatening asteroids. CRTS-II will leverage this increased data stream to discover and study objects and phenomena such as supernova and massive accreting blackholes (AGN).
The CRTS-II project follows on from the NSF-funded CRTS project which began in 2007 and discovered more than 7,000 highly variable and transient sources including significant scientific discoveries of super-luminous supernovae. CRTS-II will continue to provide a steady open stream of astronomical events, available to the entire community in real time, while further expanding the discovery space for time-domain astrophysics. The project will more than triple current sensitivity to transient objects and phenomena changing on time scales from tens of minutes to decades.
Representing CACR’s expertise in time-domain astronomy, project co-PI and research scientist Andrew Drake manages the data analysis, mining for astrophysical transients such as unusual types of supernovae, rapid transients, and outbursts of beamed active galactic nuclei (blazars). This work also employs a strategized approach aimed at enabling the serendipitous discovery of new types of objects and phenomena. Senior Computational Scientist Matthew Graham is characterizing the archival collection of 500 million sources using new statistical measures to study the broad phenomenology of behavior in the time domain.
“The project will more than triple current sensitivity to transient objects and phenomena changing on time scales from tens of minutes to decades.” says co-PI Professor George Djorgovski. More information about the project can be found on the CRTS website.