Petaflop Computing for Protein Folding

In this subproject, we are trying to map protein folding through long MD and/or HMC simulations for a Petaflop million-processor-array. We have instrumented our MD code, conducted simulations to guide the design, and built a performance model that has been validated through simulations. Current and future work involves the consideration of the mapping into a million node array, load balancing, and considering realistic simulation protocols--fast electrostatics, periodic boundary conditions, and other ensembles. We plan to multithread our code and execute it on a Tera machine. The Petaflop group at Notre Dame has a non-disclosure agreement for cooperation with IBM on the Blue Gene Petaflop computing project. We plan to compare different architectural and algorithmic approaches for MD with them, and possibly run PROTOMOL on Blue Gene prototypes, see letters of support of our collaboration by Dr. Peter Kogge, head of the Petaflop group at Notre Dame, and Dr. Mark Snir, leader of IBM's Blue Gene project.