Annual Meeting - SCEC.org
Annual Meeting - SCEC.org
Annual Meeting - SCEC.org
- TAGS
- annual
- meeting
- www.scec.org
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Group 1 – CME/PetaSHA | Poster Abstracts<br />
Community Modeling Environment (CME);<br />
Petascale Cyberfacility for Physics-Based Seismic Hazard<br />
Analysis (PetaSHA)<br />
1-019<br />
SUPPORTING <strong>SCEC</strong> PETASHAKE SIMULATIONS Cui Y, Zhu J, Chourasia A, Olsen KB,<br />
Dalguer LA, Lee K, Davis S, Day SM, Juve G, Maechling PJ, and Jordan TH<br />
We have performed a set of very-large scale earthquake simulations on the San Andreas Fault on<br />
the TeraGrid computing resources, which used more than ten millions of NSF TeraGrid allocations<br />
over the past year. The simulations are a collaborative, inter-disciplinary research program<br />
coordinated by the Southern California Earthquake Center (<strong>SCEC</strong>) that makes extensive use of<br />
large-scale, physics-based, numerical modeling of earthquake phenomena. This is part of the<br />
USGS-led public earthquake preparedness exercise called ShakeOut exercise.<br />
The <strong>SCEC</strong> AWP-Olsen based ShakeOut simulations were scientifically and computationally<br />
challenging for several reasons including: (1) the simulations were performed for a large (600km x<br />
300km x 80km) geographical volume, (2) the simulations were run at high resolution (100m spacing<br />
on a regular grid resulting in 14.4 billion mesh points), and (3) in order to simulate the earthquake<br />
rupture more realistically, source descriptions based on dynamic rupture simulations (ShakeOut-<br />
D) using about ? million subfaults was used as a primary input to the wave propagation<br />
simulations. In addition, ShakeOut simulations with kinematic source descriptions (ShakeOut-K)<br />
were also carried out for code verification purposes and to compare ground motions computed for<br />
various level of complexity.<br />
The simulations were coordinately executed on multiple TeraGrid systems, including the 504<br />
Teraflops Ranger system at Texas Advanced Computing Center (TACC), and Datastar p655 at San<br />
Diego Supercomputer Center (SDSC). Additional SDSC Datastar fat nodes p690 and Blue Gene/L<br />
were used for pre- and post-processing of these capability simulations. A large part of a hundred<br />
terabytes of simulation outputs were registered directly from TACC disks to the <strong>SCEC</strong> digital<br />
library on SAM-QFS at SDSC, managed through SDSC’s latest iRODs (integrated Rule-Oriented<br />
Data System). We aggressively optimized the iRODs transfer rate up to 135 MB/s, which is many<br />
times faster than previously recorded.<br />
Visualization for <strong>SCEC</strong> ShakeOut-K and ShakeOut-D simulations were performed on site at TACC.<br />
Visualizations were created by color mapping the data then overlaying contextual information like<br />
freeways, fault lines and topography. Comparative side by side visualization for these two sets of<br />
simulations were also created. The visualization movies are available at<br />
http://visservices.sdsc.edu/projects/scec/shakeout.<br />
1-020<br />
<strong>SCEC</strong> CME AWP-OLSEN APPLICATION ENHANCEMENTS FOR PETASCALE<br />
COMPUTING Cui Y, Kaiser TH, Zhu J, Lee K, Maechling PJ, Olsen KB, and Jordan TH<br />
The <strong>SCEC</strong> CME AWP-Olsen code has been through many optimizations over last few years to meet<br />
<strong>SCEC</strong> research requirements such as running earthquake simulations at 0.5Hz at the TeraShake<br />
scale. The recent <strong>SCEC</strong> ShakeOut exercise is performed at 1Hz, the computational requirement is<br />
16x larger than previous TeraShake simulations. The increased input size, often exceeding one<br />
terabyte in size, adds additional challenge to the scalability, memory and fault tolerance of the<br />
2008 <strong>SCEC</strong> <strong>Annual</strong> <strong>Meeting</strong> | 77