The CC-NUMA Project: Computational Chemistry on Non-Uniform Memory-access Architectures (Phase I)

Phase I has been funded by the ARC Linkage Grant LP0347178 over the years 2003-2006. Its focus has been on cc-NUMA shared memory parallel processors. Its main research platform has been an 12 CPU V1280 UltraSPARC III, an in-kind contribution from Sun Microsystems.

The project supported two Postdoctoral Fellows (Bill Clarke and Rui Yang) and two APAI PhD students (Joseph Antony and Andrew Over). Its activities have included team visits to Sun and Gaussian in 2003, a number of individual visits over 2004 and 2005, and an internship by Andrew at Sun during 2005.

Project Summary

The project will develop software tools and methods that are appropriate for current and future generations of large scale shared memory computer systems. Its purpose is to enable a more productive utilization of these architectures for scientific computation. We will focus on algorithms for solving differential equations appropriate to quantum chemistry. In particular an exciting new class of methods whose computational cost scales linearly with system size. Our goal is to develop scalable parallel implementations of these methods. If realized this will revolutionize computation, enabling first principles calculations on truly nanoscale systems, such as enzymes and molecular electronic devices.

Keywords: parallel algorithms, OpenMP, computational chemistry, computer performance evaluation, computer simulation, hardware-software co-design, non-uniform memory access shared memory multiprocessors.

Further Details of the Project

The project is broken into two themes: an Algorithm Development theme and a Performance Evaluation theme; the role of the latter is to support the former.

In brief, the Algorithm Development theme involves development of key algorithms used in the Gaussian package, their optimization on medium to large-scale NUMA parallel processors, and investigate how they can be expressed in terms of suitable extensions the OpenMP programming paradigm.

In brief, the Performance Evaluation theme involved undertaking a performance analysis of the above algorithms, with emphasis on serial and parallel memory access issues. Performance instrumentation libraries and computer simulation techniques will be used to make these analysis, with the Sun UltraSPARC III SMP being the architecture of interest. This work has included the development and extension of the Sparc-Sulima simulation tool.

More details taken from the original project proposal can be downloaded from here.

last modified: Peter Strazdins 10/06