Data-Driven, Multiscale, High Performance Applications with Visualization

Craig C. Douglas, University of Wyoming Mathematics and School of Energy Resources
Yalchin Efendiev, Texas A&M University Mathematics Department
Charles D. Hansen, University of Utah Computer Science Department

KAUST Winter Enrichment Program, January 23-27, 2010

Class: SaSuMTuW 9:30 11:30
Computer Lab: SaMW 1:00 3:30

Course Description

Numerous problems, including ones in energy, remediation, climate and weather forecasting, monitoring petro-chemical plants, and manufacturing, that until recently were solved using just static initial data on a given scale now use live data from sensor networks and multiple scales to explore different questions in real-time. The need to see solutions immediately, not eventually, has resulted in a paradigm that uses very high performance computers and visualization on a wide variety of output devices ranging from 3D immersive environments to desktops/laptops to smart phones.

The components in this course include the following:

• One application will be subsurface flow of oil, gas, or water so that either a reservoir or contaminant problem can be studied. Multiscale models at different scales will be studied and linked to each other. This will be used in developing efficient and robust multiscale data assimilation in real time.
• Parallel computing methods will be presented using a simplified library that hides communication specific details.
• Numerical methods for very high performance computers will be presented with computational libraries that hide specific details.
• Visualization techniques using SCIRun as the problem solving environment will be presented.
• Data processing from sensors to delivery to applications will be presented.

A comprehensive single example will be presented throughout the course to focus the learning experience.

In the computer laboratory sessions, students will be taught how to use SCIRun, which is a comprehensive, easy to use, modern problem solving environment. Sophisticated, already existing modules (e.g., solvers, data transmitters, visualization, and ones to create discretizations for partial differential equations) can be connected graphically (pipes) to create data flow programs for quite complicated problems without programming in a legacy programming language (e.g., C/C++/Fortran).

Who should take this course? To learn how to produce simulations using data from the field (live or archived with timestamps) in (faster than) real time on parallel computers. Be the first in your class to learn techniques that can let you control an entire refinery from your cell phone one day in the not too far off future!