Large-scale Problem Solving Using Automatic Code Generation and Distributed Visualization
Andrei Hutanu, Erik Schnetter, Werner Benger, Eloisa Bentivegna, Alex Clary, Peter Diener, Jinghua Ge, Robert Kooima, Oleg Korobkin, Kexi Liu, Frank Loffler, Ravi Paruchuri, Jian Tao, Cornelius Toole, Adam Yates, Gabrielle Allen--all Louisiana State University, Baton Rouge, LA, 70803 USA
Scientific computation today is facing many scalability challenges and issues in trying to take advantage of the latest generation compute, network and graphics hardware. We present a comprehensive approach to solving four important scalability challenges: programming productivity, scalability to a large number of processors, I/O bandwidth, and interactive visualization of large data. We describe a scenario of our proposed system applied to the field of numerical relativity, in particular the simulation of a binary black hole system. Here a solver for the governing Einstein equations is generated, executed on a large computational cluster, and its output is distributed onto a temporary distributed data server using a high-speed optical grid, and finally visualized remotely using distributed visualization methods and optical networks. This system is applicable to a large number of problems. A demonstration of this system was awarded first place in the IEEE SCALE 2009 Challenge in Shanghai, China in May 2009.
Download Article: CCT-TR-2009-11.pdf