HPA Supports IU's 100 Gbit/sec efforts at the SC11 SCinet Research Sandbox

At SC11 (November 12-18), IU will showcase a first of its kind 100 Gbit/sec production network equipped with multi-vendor OpenFlow-capable switches. HPA is supporting this entry into the SCinet Research Sandbox (SRS), "The Data Superconductor: An HPC cloud using data-intensive scientific applications, Lustre-WAN, and OpenFlow over 100Gb Ethernet," which uses the Lustre file system and cutting-edge network infrastructure to address challenges created by the exponential growth in volume of digital scientific research data.

For the SRS demonstration IU will run a series of applications using the Lustre-WAN file systems as their main storage resource. HPA has identified key applications that can benefit from such an extremely powerful network and will manage the applications as they are run across this 2000 mile 100 Gbit/sec link.

HPA works with Dr. Chen's lab on Teragrid proposal

Protein isoforms are an essential component for expanding functional complexity of the eukaryotic genomes. But systematic, proteome-scale experimental and computational characterization of protein isoforms directly at the protein/peptide level has never been reported. Dr. Chen’s lab developed a high performance computing and data analysis system to support proteomics data analysis. HPA group is working with his lab to conduct analysis of a sample data set on Quarry, and expects to find an optimal solution for running multiple parallel jobs simultaneously on larger systems. Our goal is to help his lab draw a competitive Teragrid proposal and ultimately get large allocation on Teragrid resources.

Using molecular dynamics to provide simulated data for studying neutron stars and white dwarfs

The accumulation of observations of white dwarfs and neutron stars over the last two decades has enabled astrophysicists to develop ever more precise models of these compact stellar objects. These models depend sensitively on data that is difficult or impossible to measure or observe. In some cases however, molecular dynamics simulations can step in to act as "computer experiments", and provide simulated data as a substitute.

For example, one parameter of interest is the thermal conductivity of neutron star surface material. Near the surface, densities are lower, and matter exists as a complex mixture of completely ionized atoms. There are equations for calculating the thermal conductivity of this material, which depend on a function S(q) called the "static structure factor". In terrestrial situations S(q) would be determined experimentally, but for neutron stars the required experimental (or observational) data is impossible to obtain. In this case, MD can be used as a substitute, to provide simulated data for calculating S(q).

HPA has been assisting Prof. Charles Horowitz of the IUB Department of Physics, and the IU Nuclear Theory Center in carrying out extensive MD simulations of neutron star and white dwarf material to compute thermal conductivity as well as a number of other material and transport properties. The simulations are being done on special purpose MDGRAPE-2 hardware, as well as massively parallel machines such BigRed and TeraGrid resources. We are currently carrying out an extensive set of simulations to compute the carbon-oxygen phase diagram, a key for understanding white dwarfs. Theoretical studies have resulted in widely varying proposals for the C-O phase diagram. MD hold the promise for providing the raw "computer-experimental" data necessary to decide which, if any of these is correct.

Maximum likelihood methods to determine patterns of heterozygosity and linkage disequilibrium from genomic sequences

Dr. Mike Lynch's group is developing maximum likelihood methods to analyze the patterns of heterozygosity and linkage disequilibrium found in the genome sequences of individuals, as a tool to study population biology and genome evolution. Simply put, a great deal can be learned by looking at the patterns of homozygosity and heterozygosity across a genome, when a diploid, non-inbreed organism is sequenced. However, heterozygosity must be determined by examining the individual reads (the raw data from a sequencer) for the entire genome, which is computationally intensive, as well as requiring large amounts of storage.
Our group has facilitated the use of local storage services and BigRed, and is now helping him in transition to TeraGrid, using the Data Capacitor for “portable” storage. We hope that we will also be able to further optimize the algorithms for parallel platforms.

Parallelizing the algorithm for 3D Ultrasound Segmentation

This project aims to port an algorithm for 3D Ultrasound Segmentation from Windodws onto a Linux cluster, and then parallelize the algorithm to take full advantage of multi-core processors. The long term goal is to enable real-time image segmentation for clinical use.

Describing matter in the environment of a supernova explosion

A major problem in contemporary astrophysics is to understand the mechanism of the supernova, the explosive end of massive stars. High Performance Applications enabled calculations towards this end to run in the background on unused Student Technology Center computers using IU's Condor pool, delivering tens of thousands of CPU hours to IU researchers.

Implementing 3D SPHARM surfaces registration on the Cell B.E. processor

A 3D SPHARM Surfaces Registration algorithm, which takes hours to run in MATLAB, was ported to and optimized for parallel platforms and the run time is decreased to just seconds.

PTI resources facilitate powerful new electron microscope

IU Data Capacitor and TeraGrid HPC resources help to manage massive data sets created by the advanced science instrument

Molecular dynamics simulation of dense nuclear matter on the MDGRAPE-2

MDGRAPE-2 accelerator boards help elucidate the structural, material and thermodynamic properties of dense nuclear matter in white dwarf and neutron stars.