Development of Advanced Computer Hardware and Software

Heart, Lung, And Blood Institute

Investigators

Linked publications & trials

Abstract

With the increased availability of parallel computer resources amenable to large scale scientific computing, it is essential to optimize the use of these resources. The efforts include the development of parallel computing techniques suitable for macromolecular simulation and the development of a parallel computer cluster and related software for high-efficiency simulations at low cost. Current projects include: - LoBoS: High performance computing using PC clusters - Development of parallel QM/MM methods - Development of bioinformatics tools for PC clusters - Development and evaluation of parallel algorithms for molecular dynamics - Development and support of parallel CHARMM - Development of Latency-tolerant algorithms for Parallel Computing The LoBoS (Lots of Boxes on Shelves), LoBoSII (and pending LoBoSIII/Biowulf II) supercomputers have been designed and constructed using commodity PCs. They provide a greater than 10-fold improvement in price/performance when compared with the traditional supercomputer vendor's offerings. The initial approach taken for LoBoS had been very successful for certain applications with 128 processors and multiple fast Ethernet connections per node. LoBoSII is a substantial improvement in scalability due to a 5-fold improvement in bandwidth using Gigabit Ethernet and customizations of the TCP stack in the Linux kernel have reduced latency by half. LoBoSIII will have 256 Pentium III 866MHz processors and a fully switched Myrinet network for half of the computers. The LoBoS Project has produced the most capable computational system at the NIH for most applications involving computational chemistry tools. It has opened up a new realm of high performance computing which continues to drive the cost down while improving reliability through the use of loosely-coupled clusters. The classes of applications that execute efficiently on clusters have been characterized and efforts to better support latency-and bandwidth-tolerant algorithms in hardware are underway. The main effort in developing new bioinformatics tools for the PC cluster environemnt is the development of a new architecture for the integration of disparate biological databases and for the distribution of compute-intensive tasks. Genomic and proteomic applications such as FASTA and BLAST are being used to optimize their execution environment within clusters by an order of magnitude. Clusters are uniquely suited to large database applications since each compute node has its own disk (maximizing IO) and its workload may be configured to be independent of other nodes (maximizing efficiency).