Relativistic Astrophysics on the SiCortex Architecture
Erik Schnetter, Steven R. Brandt
Numerical simulations of astrophysical systems are often the only way to gain insight into observational data, since observations are limited by the systems' large distances, and experiments are often impossible due to the systems' extreme physical conditions. This is especially true in relativistic astrophysics where the complex features of black holes or neutron stars can not be analyzed in other ways. Such simulations can use formidable amounts of computing resources, and parallel high performance computing systems are the only viable way to perform these simulations.
New parallel systems are projected to provide several Petaflop/s sustained performance for scientific applications, using new hardware architectures and new levels of parallelism to achieve this. Using such systems will likely require the use of higher levels of abstractions such as provided by the Cactus software framework, and/or require a shift to new programming models such as provided by ParalleX.
The SiCortex architecture provides a mixture of traditional features found on today's Linux clusters and novel properties that are also likely to be found on future systems. It can thus serve as stepping stone to help developers research algorithms and tools to ready their applications for Petascale computing.
Download Article: CCT-TR-2009-4.pdf