SGER:Performance Analysis of the BlueGene Class of Machines via the ASTRO-FOLD Protein Structure Prediction Framework

Princeton University, Princeton NJ

Investigators

Abstract

Abstract, CNS-0401635 The project will port the Astro-Fold first principles methodology for protein structure prediction to the BlueGene class of machines with primary objectives: (i) to conduct a performance analysis of the BlueGene architecture, and (ii) to validate the architecture's unique features of concurrency between computation and communication. The project will develop parameter representation of the distributed system and will perform distributed computing studies with different number of nodes so as to assess and understand the performance of the network delays on the computations, the message passing and the communications overhead. The project will also investigate and assess the performance of the BlueGene class of machines not only on the individual levels of distributed computations (i.e, the helical prediction, the loop prediction, and the three-dimensional structure prediction), but also on the combined distributed computing performance of all levels and hierarchy in the distributed computations. This will be of major significance for the BlueGene class of machines and for the Astro-Fold methodology.