Exploring Tuning Strategies for Quantum Chemistry Computations

Effective execution of applications using a parallel computing environment that share resources such as network bandwidth, I/O and main memory require that some sort of adaptive mechanism be put in place which enables efficient usage of these resources. The adaptation adjusts the most computationally intensive part of the application thus leading to an efficient execution. General Atomic and Molecular Electronic Structure (GAMESS), used for ab-initio molecular quantum chemistry calculations, uses NICAN for dynamically making adaptations so as to improve the application performance in heavy load conditions. The adaptation mechanism uses the iteration time of each SCF iteration to make a decision. The advantage of such a mechanism is the ability to modify the application execution in a very simplistic yet effective manner. In this work, we have explored methods to expand this adaptation mechanism by analyzing the GAMESS performance through the use of fine-grained data.