The International Conference for High Performance Computing, Networking, Storage and Analysis
Fine-grained Gathering of Scientific Data in QMCPack Simulations on Titan.
Student: Stephen N. Herbein (University of Delaware)
Supervisor: Scott Klasky (Oak Ridge National Laboratory)
Abstract: Traditional petascale applications, such as QMCPack, can scale their computation to completely utilize modern supercomputers like Titan, but cannot scale their I/O. Because of this lack of scalability, scientists cannot save data at the granularity level they need to enable scientific discovery.
In this work, we tackle the problem of increasing the granularity of data collected from QMCPack simulations without increasing I/O overhead or compromising the simulation’s scalability. Our solution relies on the redesign of the QMCPack algorithms to gather fine-grained information and the integration of the ADIOS API to help select effective I/O methods without major code changes.
Results presented in the poster outline how we can increase the quality of the scientific knowledge extracted from Titian simulations of QMCPack while keeping the I/O overhead below 10%.