Pittsburgh Supercomputing Center Research Articles

Abstract 5109: National cyberinfrastructure and bioinformatic analysis support available to the cancer research community

Abstract National Center for Genomic Analysis Support (NCGAS) is an NSF-funded center whose goal is to help biologists optimize their bioinformatic pipelines to best address their research questions. To help reach these goals, NCGAS provides a range of services based on the researchers needs. We currently host gateways, Trinity Galaxy and Genepattern, with built-in workflows to support cancer researchers who have little background in command line work. Both these gateways are projects funded by the Information Technologies in Cancer Research (ITCR) program of NIH, NCI, to aid in the genomic and transcriptomic characterization of cancers. NSF-funded researchers comfortable in command line can gain access to IU cyberinfrastructure (HIPAA compliant), to run or develop bioinformatic workflows. Bioinformatic packages are already installed system wide, with reference databases, and new programs are always added upon user’s request. NCGAS is also an ambassador of Extreme Science and Engineering Discovery Environment (XSEDE) resources, providing access to other HIPAA compliant national cyberinfrastructure (Texas Advanced Computing Center, Pittsburgh Supercomputing Center, etc.), available to the entire research community. XSEDE resources include the Jetstream cloud, based at IU and TACC, allowing researchers to use preconfigured virtual machines (VMs) already setup, create custom VM environments for their own work, and to share with classes and collaborators. Data can be transferred and managed between all of these resources securely with already available Globus endpoints which allows transfer for terabytes of data quickly. Additionally, NCGAS hosts workshops to help researchers get started. These training events can be scheduled online or at your local university upon request. NCGAS currently supports projects working on transcriptome, genome-level assembly, phylogenetics, metagenomics, and meta-transcriptomics analysis through providing a range of compute resources that best serve the research project. Citation Format: Bhavya Papudeshi, Carrie Ganote, Sheri Sanders, Thomas G. Doak. National cyberinfrastructure and bioinformatic analysis support available to the cancer research community [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 5109.

Deciphering the ‘fuzzy’ interaction of FG nucleoporins and transport factors using SANS

The largely intrinsically disordered phenylalanine-glycine-rich nucleoporins (FG Nups) underline a selectivity mechanism, which enables the rapid translocation of transport factors (TFs) through the nuclear pore complexes (NPCs). Conflicting models of NPC transport have assumed that FG Nups undergo different conformational transitions upon interacting with TFs. To selectively characterize conformational changes in FG Nups induced by TFs we performed small-angle neutron scattering (SANS) with contrast matching. Conformational ensembles derived SANS data indicate an increase in the overall size of FG Nups is associated with TF interaction. Moreover, the organization of the FG motif in the interacting state extends prior experimental analyses defining that FG motifs undergo conformational restriction upon interacting with TFs. These results provide structural insights into a highly dynamic interaction and illustrate how functional disorder imparts rapid and selective FG Nup – TF interactions. Support or Funding Information This work was supported by NIH GM117212(DC) and GM109824. This research used resources at the High Flux Isotope Reactor, a DOE Office of Science User Facility operated by the Oak Ridge National Laboratory. Simulations used the Anton special-purpose supercomputer provided by the National Resource for Biomedical Supercomputing (NRBSC), the Pittsburgh Supercomputing Center (PSC), and the Biomedical Technology Research Center for Multiscale Modeling of Biological Systems (MMBioS) through Grant P41GM103712-S1 from the National Institutes of Health. D. E. Shaw Research generously made the Anton machine available. Simulations also used the XSEDE facilities supported by NSF ACI-105375. (A) Fit of best-performing sub-ensembles.(B) Plots of the Rg distribution compared with the initial pool of TraDES models (Black) for the SANS data indicated.(C) Weighted average Cα-Cα distance maps.(D) Conformers of the best-performing subensemble. Left, [2H]-FSFG-K, middle [2H]-FSFG-K + NTF2 (0.6 mM) (TraDES pool), (right) [2H]-FSFG-K + NTF2 (0.6 mM) with selection from the MD pool. This abstract is from the Experimental Biology 2018 Meeting. There is no full text article associated with this abstract published in The FASEB Journal.

Translocation of Potassium in Initiating Reset of Monoamine Transporter Proteins from Microsecond Molecular Dynamics Simulations

Our intent is to characterize the complete transport mechanism of the serotonin transporter (SERT) computationally, employing both a single bilayer and dual bilayer system to investigate this phenomenon. The work presented here represents the data collected from our most recent dual bilayer system from 178,044 atom simulations after 1 µs of simulation time. Herein, we utilize the human SERT protein crystal structure (PDB 5I6X), embedded in POPE lipid membrane. Ions were added to either side of the membrane at physiological concentrations, maintaining the membrane potential at −67 mV, representative of resting potential. Potential was determined using the Goldman-Hodgkin-Katz equation. Systems were run on both XSEDE and Pittsburgh Supercomputing Center resources. After 1 µs molecular dynamics simulations using the CHARMM force field, we observe translocation of potassium to the central substrate binding site in both SERT and the dopamine transporter (DAT). In both cases, potassium is interacting with a conserved aspartic acid residue in the pocket (D98 in SERT and D79 in DAT). The translocation of potassium to the central binding pocket after 1 µs supports experimental data that suggests potassium is necessary for reset of SERT. Simulations are currently running to characterize the complete transport mechanism. A more complete understanding of the transport process can aid in drug discovery and design for treatments of disorders associated with these proteins, such as depression and related disorders.

Asymmetric Membrane Models for the PM And TGN of Yeast, An All-Atom Molecular Dynamics Study

We previously examined symmetric membrane models for the plasma (PM) and trans-Golgi-network (TGN) membranes of Saccharomyces cerevisiae (yeast) (Biochem 54:6852-6861). Although diverse in lipid nature and reflecting sterol compositions characteristic to those organelles, our models were not an accurate enough representation as they lacked sphingolipids (SM), relevant for membrane structure and dynamics. Once SM were parametrized for the CHARMM 36 lipid force field (BJ, 107:134-145), we included them in the new models and added lipid composition asymmetry to better reflect the complexity of these bilayers. The new models include inositol phosphoceramide (IPC) and mannose-(inositol-P)2-ceramide M(IP)2C lipids, both present in the non-cytosolic leaflets of the PM and TGN. We present a comparison study between the previous PM model and the more complex ones using common membrane properties such as surface area per lipid, bilayer thickness, deuterium order parameters, lipid diffusion constants, and lipid cluster formation. The last three properties will serve to investigate lipid phase coexistence and leaflet coupling in membrane order and diffusion. Our simulation trajectories were at least 2µs long for each of the three models (old-PM, PM, and TGN), for a total of 15 µs of simulation data run in the Anton machine at the Pittsburgh Supercomputing Center.

Scalable 3D hybrid parallel Delaunay image-to-mesh conversion algorithm for distributed shared memory architectures

In this paper, we present a scalable three-dimensional hybrid parallel Delaunay image-to-mesh conversion algorithm (PDR.PODM) for distributed shared memory architectures. PDR.PODM is able to explore parallelism early in the mesh generation process thanks to the aggressive speculative approach employed by the Parallel Optimistic Delaunay Mesh generation algorithm (PODM). In addition, it decreases the communication overhead and improves data locality by making use of a data partitioning scheme offered by the Parallel Delaunay Refinement algorithm (PDR). PDR.PODM supports fully functional volume grading by creating elements with varying size. Small elements are created near boundary or inside the critical regions in order to capture the fine features while big elements are created in the rest of the mesh. We tested PDR.PODM on Blacklight, a distributed shared memory (DSM) machine in Pittsburgh Supercomputing Center. For the uniform mesh generation, we observed a weak scaling speedup of 163.8 and above for up to 256 cores as opposed to PODM whose weak scaling speedup is only 44.7 on 256 cores. PDR.PODM scales well on uniform refinement cases running on DSM supercomputers. The end result is that PDR.PODM can generate 18 million elements per second as opposed to 14 million per second in our earlier work. The varying size version sharply reduces the number of elements compared to the uniform version and thus reduces the time to generate the mesh while keeping the same fidelity.

Two-level locality-aware parallel Delaunay image-to-mesh conversion

The Locality-Aware Parallel Delaunay (LAPD) method exploits concurrency at two different levels of granularities (data region and cavity) and employs data partitioning to improve data locality; as a result LAPD scales well on Distributed Shared Memory computers with about 200 cores. In this paper, we propose a three dimensional two-level Locality-Aware Parallel Delaunay image-to-mesh conversion algorithm (LAPD). The algorithm exploits two levels of parallelism at different granularities: coarse-grain parallelism at the region level (which is mapped to a node with multiple cores) and medium-grain parallelism at the cavity level (which is mapped to a single core). We employ a data locality-aware mesh refinement process to reduce the latency caused by the remote memory access. We evaluated LAPD on Blacklight, a cache-coherent NUMA distributed shared memory (DSM) machine in the Pittsburgh Supercomputing Center, and observed a weak scaling efficiency of almost 70% for roughly 200 cores, compared to only 30% for the previous algorithm, Parallel Optimistic Mesh Generation algorithm (PODM). To the best of our knowledge, LAPD exhibits the best scalability for parallel Delaunay mesh generation algorithms running on NUMA DSM supercomputers.

Scalable 3D Hybrid Parallel Delaunay Image-to-Mesh Conversion Algorithm for Distributed Shared Memory Architectures

In this paper, we present a scalable three dimensional hybrid parallel Delaunay image-to-mesh conversion algorithm (PDR.PODM) for distributed shared memory architectures. PDR.PODM is able to explore parallelism early in the mesh generation process because of the aggressive speculative approach employed by the Parallel Optimistic Delaunay Mesh generation algorithm (PODM). In addition, it decreases the communication overhead and improves data locality by making use of a data partitioning scheme offered by the Parallel Delaunay Refinement algorithm (PDR). PDR.PODM utilizes an octree structure to decompose the initial mesh and to distribute the bad elements to different octree leaves (subregions). A set of independent subregions are selected and refined in parallel without any synchronization among them. In each subregion, a group of threads is assigned to insert or delete multiple points based on the refinement rules offered by PODM. We tested PDR.PODM on Blacklight, a distributed shared memory (DSM) machine in the Pittsburgh Supercomputing Center, and observed a weak scaling speedup of 163.8 and above for up to 256 cores as opposed to PODM whose weak scaling speedup is only 44.7 on 256 cores. The end result is that we can generate 18 million elements per second as opposed to 14 million per second in our earlier work. To the best of our knowledge, PDR.PODM exhibits the best scalability among parallel guaranteed quality Delaunay mesh generation algorithms running on DSM supercomputers.

Structure of Transmembrane Pores Stabilized by Antimicrobial Peptides Magainin and PGLa

Antimicrobial peptides are found in many organisms as part of their defense system against bacterial infections. They have similar structural and functional features: most consist of amphiphilic helices that bind to a membrane and disrupt it by diverse methods. Among these peptides, magainin 2 and PGLa are found in the frog skin and exhibit synergistic effects in lipid bilayer disruption by the formation of transmembrane pores [1]. Experimental methods, such as solid-state NMR, have shed light on the relative orientation of magainin 2 and PGLa on membranes with different lipid composition [2,3], but there is a general lack of information on the structure of the pore and the specific interactions that lead to its stabilization. In the present work, we studied the structure and dynamics of transmembrane pores formed by magainin 2 and magainin 2/PGLa (2:2) tetramers by all-atom molecular dynamics simulations performed at the Anton supercomputer (Pittsburgh Supercomputing Center). For both systems we observed stabilization of a pore. The 9-μs simulation time allows a detailed analysis of its structure and properties, the role of the lipids surrounding it, and the relative orientation of magainin 2 and PGLa in the membrane.

High quality real-time Image-to-Mesh conversion for finite element simulations

In this paper, we present a parallel Image-to-Mesh Conversion (I2M) algorithm with quality and fidelity guarantees achieved by dynamic point insertions and removals. Starting directly from an image, its implementation is capable of recovering the isosurface and meshing the volume with tetrahedra of good shape. Our tightly-coupled shared-memory parallel speculative execution paradigm employs carefully designed contention managers, load balancing, synchronization and optimizations schemes. These techniques are shown to boost not only the parallel but also the single-threaded efficiency of our code. Specifically, our single-threaded performance is faster than both CGAL and TetGen, the state of the art sequential open source meshing tools we are aware of. The effectiveness of our method is demonstrated on Blacklight, the Pittsburgh Supercomputing Center’s cache-coherent NUMA machine. We observe a more than 82% strong scaling efficiency for up to 64 cores, and a more than 82% weak scaling efficiency for up to 144 cores, reaching a rate of more than 14.3 million elements per second. This is the fastest 3D Delaunay mesh generation and refinement algorithm, to the best of our knowledge.

GPCR Activation on the Microsecond Timescale in MD Simulations

G-protein coupled receptors (GPCRs) play a central role in signal transduction and consequently represent a major drug target. Structural similarities suggest a common working mechanism, which is typically deduced from crystal structures of activation intermediates. The latter have become available recently, in particular for the light sensing GPCR rhodopsin and other members of the rhodopsin-like GPCR family. These crystal structures provide snapshots along the activation pathway, however, the transitions between those remain speculative.Time resolved spectroscopy is able to provide crucial insights into dynamic details. However, another promising route to study receptor dynamics was opened by the recent occurrence of special purpose computers for molecular dynamics (MD) simulations, which allow to observe conformational transitions related to receptor activation/deactivation directly in unbiased MD simulations on timescales of tens of microseconds.The activation kinetics for GPCRs, such as rhodopsin, typically feature time constants on the order of milliseconds, which is still beyond reach for direct simulation. However, here we present microsecond simulations for a special case, squid rhodopsin from Todarodes pacificus. While it shares many structural features with its bovine counterpart, activation occurs with a time constant of 12.5 microseconds at room temperature. We analyze simulations of the light activated receptor spanning up to 20 microseconds, allowing us to identify activation events and focus specifically on the interactions between the retinal ligand and the receptor, that trigger the activation process.This work was supported by the German Academy of Sciences Leopoldina (fellowship to M.H.) and the NSF (grant CHE-0750175). We are grateful for an allocation of computer time on Anton at the Pittsburgh Supercomputing Center (supported by the NIH).

Understanding Substrate Unbinding from the Sodium-Galactose Co-Transporter vSGLT based on 16 Microseconds of Molecular Simulation

We report the results from 16 microseconds of molecular dynamics (MD) simulations carried out on vSGLT using the Anton supercomputer at the Pittsburgh Supercomputing Center. We observed multiple galactose unbinding events as well as instances in which the energizing sodium ion unbind and then rebind to the putative sodium binding site. Using umbrella sampling we calculated the potential of mean force (pmf) for all galactose unbinding trajectories. We will discuss the allosteric interaction between the ion and substrate and the potential for one to control the release of the other via the inner and outer gates.

Microscopic Origin of Gating Current Fluctuations in a Potassium Channel Voltage Sensor

Voltage-sensing domains (VSDs) regulate the opening of voltage-dependent ion channels in response to changes in membrane electrical potential. They are comprised of four transmembrane segments (S1 through S4) and their overall architecture is conserved from Archaea to mammals. VSD charge displacements within the membrane electric field after a sudden change in membrane potential are measured in electrophysiology experiments as capacitive currents, termed gating currents, which precede ion conduction. The elementary charge motions that give rise to the gating current appear as shot-like events in high-bandwidth recordings of gating current fluctuations, but cannot be observed directly due to limitations of present recording technology. To characterize the conformational dynamics of the VSD under an applied membrane potential, we performed all-atom MD simulations of the isolated Kv VSD from Aeropyrum pernix (KvAP) in a 1-palmitoyl-2-oleoyl-phosphatidylcholine (POPC) bilayer in excess water under a constant electric field on the timescale (10-μs) expected for elementary gating charge transitions. The results suggest that shot-like events of gating current are associated with the water-catalyzed rearrangement of salt bridges between the S4 arginines and a set of conserved acidic side chains on the S1-S3 transmembrane segments in the hydrated interior of the VSD. Research supported by NIH grants GM74637 and GM86685 and NSF grant CHE-0750175. The computations were carried out on the Anton computer at the National Resource for Biomedical Supercomputing at the Pittsburgh Supercomputing Center, which is funded by NIH grant RC2GM093307.

Proceedings of the 2011 MidSouth Computational Biology and Bioinformatics Society (MCBIOS) Conference

The Eighth Annual Conference of the MidSouth Computational Biology and Bioinformatics Society (MCBIOS’2011) was held in College Station, Texas on April 1-2, 2011. The Conference General Chair was Ulisses Braga-Neto, the MCBIOS President for the 2010-2011 term, from Texas A&M University. There were nearly 200 registrants and 140 abstracts were submitted, divided into 48 oral presentation abstracts and 92 poster session abstracts. In addition, participants attended talks by very distinguished Keynote speakers. Joan W. Bennett, from Rutgers University and Member of the National Academy of Science, presented the talk “Chromosomal Composition and Computational Competence;” Donald Geman, from The Johns Hopkins University and co-inventor of the Gibbs Sampler and Random Forest Classifiers, lectured on “Measuring Network Regulation and Differential Expression by Rank Conservation;” while John Weinstein, Chair of the Department of Bioinformatics and Computational Biology at the University of Texas M.D. Anderson Cancer Center, talked on “Personalizing Cancer Medicine in the Era of Next-Generation Sequencing: Omics and Informatics.” The conference also benefited from invited dinner and lunch speakers, who gave informal, highly informative and entertaining talks. Edward R. Dougherty, from Texas A&M University and Director of Computational Biology at the Translational Genomics Institute, asked the audience “Is Biological Science Delightful?” whereas Ernesto Marques, from the Center for Vaccine Research at the University of Pittsburgh, gave the talk “Activation of the Complement System in Dengue Infection: Opportunities for Computational Modeling.” Participants also had the opportunity to attend hands-on workshops on NCBI tools, presented by Peter Cooper from NCBI/NIH, and on protein evolution, presented by Hugh Nicholas and Troy Wymore, from the Pittsburgh Supercomputing Center. The winners of conference awards were: Best Oral Presentations (students): First place: Suzanne Matthews, Texas A&M University Second place (3-way tie): Lori Dalton, Texas A&M University Second place: Shelton Griffith, Oak Ridge National Lab Second place: Winston Haynes, Hendrix College Best Oral Presentations (Post-Doctoral fellows): Yan Li, NCTR Fan Zhang, IUPUI Best Poster (Computation): First place: Tianchuan Du, Southern University Second place: Christopher Cathey, Jackson State University Third place: Ralph Crosby, Texas A&M University Best Poster (Biology): First place: Awantika Singh, UALR/UAMS Second place: Mohammed Shahrokh Esfahani, Texas A&M University Third place: Fang-Han Hsu, Texas A&M University MCBIOS Outstanding Service Award: Dr. Jonathan D. Wren, Oklahoma Medical Research Foundation

Lattice QCD simulations on big cats, sea monsters and clock towers

We present details of lattice QCD computations we are performing on the Cray XT series of computers, from BigBen – an XT3 hosted at the Pittsburgh Supercomputing Center (PSC) – through Jaguar (XT4) and Kraken (XT5) – which are hosted at the National Center for Computational Science (NCCS) and the National Institute of Computational Science (NICS), respectively, at Oak Ridge National Laboratory (ORNL). We discuss algorithmic tuning to make the computation more efficient and present some recent results.

QCMPI: A parallel environment for quantum computing

QCMPI is a quantum computer (QC) simulation package written in Fortran 90 with parallel processing capabilities. It is an accessible research tool that permits rapid evaluation of quantum algorithms for a large number of qubits and for various “noise” scenarios. The prime motivation for developing QCMPI is to facilitate numerical examination of not only how QC algorithms work, but also to include noise, decoherence, and attenuation effects and to evaluate the efficacy of error correction schemes. The present work builds on an earlier Mathematica code QDENSITY, which is mainly a pedagogic tool. In that earlier work, although the density matrix formulation was featured, the description using state vectors was also provided. In QCMPI, the stress is on state vectors, in order to employ a large number of qubits. The parallel processing feature is implemented by using the Message-Passing Interface (MPI) protocol. A description of how to spread the wave function components over many processors is provided, along with how to efficiently describe the action of general one- and two-qubit operators on these state vectors. These operators include the standard Pauli, Hadamard, CNOT and CPHASE gates and also Quantum Fourier transformation. These operators make up the actions needed in QC. Codes for Grover's search and Shor's factoring algorithms are provided as examples. A major feature of this work is that concurrent versions of the algorithms can be evaluated with each version subject to alternate noise effects, which corresponds to the idea of solving a stochastic Schrödinger equation. The density matrix for the ensemble of such noise cases is constructed using parallel distribution methods to evaluate its eigenvalues and associated entropy. Potential applications of this powerful tool include studies of the stability and correction of QC processes using Hamiltonian based dynamics. Program summary Program title: QCMPI Catalogue identifier: AECS_v1_0 Program summary URL: http://cpc.cs.qub.ac.uk/summaries/AECS_v1_0.html Program obtainable from: CPC Program Library, Queen's University, Belfast, N. Ireland Licensing provisions: Standard CPC licence, http://cpc.cs.qub.ac.uk/licence/licence.html No. of lines in distributed program, including test data, etc.: 4866 No. of bytes in distributed program, including test data, etc.: 42 114 Distribution format: tar.gz Programming language: Fortran 90 and MPI Computer: Any system that supports Fortran 90 and MPI Operating system: developed and tested at the Pittsburgh Supercomputer Center, at the Barcelona Supercomputer (BSC/CNS) and on multi-processor Macs and PCs. For cases where distributed density matrix evaluation is invoked, the BLACS and SCALAPACK packages are needed. Has the code been vectorized or parallelized?: Yes Classification: 4.15 External routines: LAPACK, SCALAPACK, BLACS Nature of problem: Analysis of quantum computation algorithms and the effects of noise. Solution method: A Fortran 90/MPI package is provided that contains modular commands to create and analyze quantum circuits. Shor's factorization and Grover's search algorithms are explained in detail. Procedures for distributing state vector amplitudes over processors and for solving concurrent (multiverse) cases with noise effects are implemented. Density matrix and entropy evaluations are provided in both single and parallel versions. Running time: Test run takes less than 1 minute using 2 processors.

Bottom-Up Construction and 2:1 Balance Refinement of Linear Octrees in Parallel

In this article, we propose new parallel algorithms for the construction and 2:1 balance refinement of large linear octrees on distributed memory machines. Such octrees are used in many problems in computational science and engineering, e.g., object representation, image analysis, unstructured meshing, finite elements, adaptive mesh refinement, and N-body simulations. Fixed-size scalability and isogranular analysis of the algorithms using an MPI-based parallel implementation was performed on a variety of input data and demonstrated good scalability for different processor counts (1 to 1024 processors) on the Pittsburgh Supercomputing Center's TCS-1 AlphaServer. The results are consistent for different data distributions. Octrees with over a billion octants were constructed and balanced in less than a minute on 1024 processors. Like other existing algorithms for constructing and balancing octrees, our algorithms have $\mathcal{O}(N\log N)$ work and $\mathcal{O}(N)$ storage complexity. Under reasonable assumptions on the distribution of octants and the work per octant, the parallel time complexity is $\mathcal{O}(\frac{N}{n_p}\log(\frac{N}{n_p})+n_p\log n_p)$, where $N$ is the size of the final linear octree and $n_p$ is the number of processors.

An Introduction to Simulation and Visualization of Biological Systems at Multiple Scales: A Summer Training Program for Interdisciplinary Research

Advances in biomedical research require a new generation of researchers having a strong background in both the life and physical sciences and a knowledge of computational, mathematical, and engineering tools for tackling biological problems. The NIH-NSF Bioengineering and Bioinformatics Summer Institute at the University of Pittsburgh (BBSI @ Pitt; www.ccbb.pitt.edu/bbsi) is a multi-institutional 10-week summer program hosted by the University of Pittsburgh, Duquesne University, the Pittsburgh Supercomputing Center, and Carnegie Mellon University, and is one of nine Institutes throughout the nation currently participating in the NIH-NSF program. Each BBSI focuses on a different area; the BBSI @ Pitt, entitled "Simulation and Computer Visualization of Biological Systems at Multiple Scales", focuses on computational and mathematical approaches to understanding the complex machinery of molecular-to-cellular systems at three levels, namely, molecular, subcellular (microphysiological), and cellular. We present here an overview of the BBSI @ Pitt, the objectives and focus of the program, and a description of the didactic training activities that distinguish it from other traditional summer research programs. Furthermore, we also report several challenges that have been identified in implementing such an interdisciplinary program that brings together students from diverse academic programs for a limited period of time. These challenges notwithstanding, presenting an integrative view of molecular-to-system analytical models has introduced these students to the field of computational biology and has allowed them to make an informed decision regarding their future career prospects.

Building the TeraGrid

On 1 October 2004, the most ambitious high-performance Grid project in the United States-the TeraGrid-became fully operational. Resources at nine sites--the San Diego Supercomputer Center, the California Institute of Technology, the National Center for Supercomputing Applications, the University of Chicago/Argonne National Laboratory, Pittsburgh Supercomputing Center, Texas Advanced Computing Center, Purdue University, Indiana University and Oak Ridge National Laboratory-were joined via an ultra-fast optical network, unified policies and security procedures and a sophisticated distributed computing software environment. Funded by the National Science Foundation, the TeraGrid enables scientists and engineers to combine distributed, multiple data sources with computation at any of the sites or link massively parallel computer simulations to extreme-resolution visualizations at remote sites. A single shared utility lets multiple resources be easily leveraged and provides improved access to advanced computational capabilities. One of the demonstrations of this new model for using distributed resources, Teragyroid, linked the infrastructure of the TeraGrid with computing resources in the United Kingdom via a transatlantic data fibre link. Once connected, the software framework of the RealityGrid project was used to successfully explore lattice-Boltzmann simulations involving lattices of over one billion sites.

A study of I/O methods for parallel visualization of large-scale data

This paper presents two parallel I/O methods for the visualization of time-varying volume data in a high-performance computing environment. We discuss the interplay between the parallel renderer, I/O strategy, and file system, and show the results of our study on the performance of the I/O strategies with and without MPI parallel I/O support. The targeted application is earthquake modeling using a large 3D unstructured mesh consisting of one hundred millions cells. Our test results on the HP/Compaq AlphaServer operated at the Pittsburgh Supercomputing Center demonstrate that the I/O methods effectively remove the I/O bottlenecks commonly present in time-varying data visualization, and therefore help significantly lower interframe delay. This high-performance visualization solution allows scientists to explore their data in the temporal, spatial, and visualization domains at high resolution. Such new explorability, likely not presently available to most computational science groups, will help lead to many new insights into the modeled physical and chemical processes.

Predicting the performance of synchronous discrete event simulation

We develop a model to predict the performance of synchronous discrete event simulation. Our model considers the two most important factors for the performance of synchronous simulation: load balancing and communication. The effect of load balancing in a synchronous simulation is computed using probability distribution models. We derive a formula that computes the cost of synchronous simulation by combining a communication model called LogGP and computation granularity. Even though the formula is simple, it is effective in capturing the most important factors for the synchronous simulation. The formula helps us to predict the maximum speed up achievable by synchronous simulation. In order to examine the prediction model, we have simulated several large ISCAS logic circuits and a simple PCS network simulation on an SGI Origin 2000 and Terascale Computing System (TCS) at the Pittsburgh Supercomputing Center. The results of the experiment show that our performance model accurately predicts the performance of synchronous simulation. The performance model developed is used to analyze the effect of several factors that may improve the performance of synchronous simulation. The factors include problem size, load balancing, granularity, communication overhead, and partitioning.