NIST Team Awarded One Million Hours on NASA Supercomputer
A team of researchers from NIST has been awarded 1,000,000 CPU hours on the Columbia supercomputer at NASA Ames Research Center. The allocation is one of four awards of supercomputer time given out by NASA in a peer-reviewed competition for grand challenge computational science projects led by external researchers. The successful NIST proposal was submitted by William George and Judith Terrill of ACMD, along with Nicos Martys and Edward Garboczi of The NIST Engineering Laboratory. Entitled "Modeling the Rheological Properties of Suspensions: Application to Cement Based Materials" the proposal stems from a long-term ACMD/EL collaboration on high performance computer modeling of cement and concrete systems. This award of such a highly sought after computing resource highlights the national importance and technical challenges associated with concrete rheology, which has a tremendous impact on the construction of concrete structures. Concrete construction is a $110B per year sector of the US economy.
The team will use NASA's supercomputer to study the flow, dispersion and merging of dense suspensions composed of rigid bodies having a wide range of size and shape under a variety of flow conditions. Access to the NASA machine will allow modeling at a level and range impossible with existing computing facilities available at NIST. Current modeling of suspensions at NIST facilities has been limited to a few thousand particles and a factor of five to ten in particle size range. Utilization of NASA's Columbia system will provide the capability to simulate suspensions an order of magnitude larger in the number of inclusions and size range. The new realism of these models will significantly improve the scientific basis for prediction and measurement of the flow properties of concrete.
Columbia is a 10,240-CPU system based on SGI's NUMAflex architecture. The system is comprised of 20 SGI Altix 3700 superclusters, each with 512 Intel Itanium 2 processors (rated at 1.5 GHz). Each supercluster features 1 terabyte of memory with global shared memory access, for a total of 20 terabytes of memory system-wide. Columbia was put into production in June 2004.