Tutorials

Presenters: Lori Diachin (Lawrence Livermore National Laboratory), Mark Shephard (Rensselaer Polytechnic Institute) and Tim Tautges (Argonne National Laboratory)

Advanced simulations need advanced software tools to manage the complexities associated with sophisticated geometry, mesh, and field manipulation tasks, particularly as computer architectures move toward the petascale. The Center for Interoperable Technologies for Advanced Petascale Simulations (ITAPS) is creating interoperable and interchangeable mesh, geometry, and field manipulation services that are of direct use to applications. The premise of our technology is that such services can be provided as libraries that can be used with minimal intrusion into application codes. In this tutorial, we give an overview and introduction to the ITAPS philosophy as well as several examples of how our technology has been used to impact existing application codes. We briefly describe several services that are available to application scientists including mesh quality improvement, front tracking, mesh adaptation, and dynamic partitioning. Underlying these services are a common data model and interfaces that are key to providing uniform access to all ITAPS tools. We will teach attendees basic use of the ITAPS interfaces through simple examples and provide advice on best practices for use of ITAPS software. We conclude the tutorial with a description of how to build your own ITAPS compliant implementation and detailed information on obtaining ITAPS software.

Presenters: Rob Ross and Rob Latham (Argonne National Laboratory)

I/O on HPC systems is a black art. This tutorial sheds light on the state-of-the-art in parallel I/O and provides the knowledge necessary for attendees best to leverage I/O resources available to them. We cover the entire I/O software stack from parallel file systems at the lowest layer, to intermediate layers (such as MPI-IO), and finally high-level I/O libraries (such as HDF-5). We emphasize ways to use these interfaces that result in high performance, and benchmarks on real systems are used to show real-world results. In this short version of the tutorial we will focus on the upper layers of the I/O stack, covering POSIX I/O, MPI-IO, Parallel netCDF, and HDF5. We will discuss interface features, show code examples, and describe how application calls translate into Parallel File System operations.

Presenters: Ian Foster and Jennifer Schopf (Argonne National Laboratory and the University of Chicago)

Do your computations or experiments generate large quantities of data that you then need to move rapidly and reliably to remote locations? Do you want to be able to create “services” that allow remote users to analyze your data and/or run application programs? Are you already doing such things, but need tools that can diagnose causes of poor performance and unexpected failures? In this tutorial, you will learn about the tools that participants in the Center for Enabling Distributed Petascale Science (CEDPS) are creating for these and related tasks. You’ll also have the opportunity to engage in a discussion about your problems and how CEPDS can help address those problems.

Presenters: James Osborn (Boston University) and Andrew Pochinsky (MIT)

The SciDAC-1 and SciDAC-2 Lattice Infrastructure Projects have developed a powerful layered software environment that enables physicists to address a broad range of fundamental problems in lattice field theory while making both efficient use of their time and efficient use of terascale computers.   The QCD Data Parallel Interface (QDP/C) is a C-based interface that utilizes  optimized linear algebra routines, message passing routines, and I/O routines that have been highly optimized to run on all the DOE  platforms available for large-scale lattice calculations.  
This tutorial will introduce a new user to the basic use of QDP/C through running and modifying hands-on examples of basic calculations such as generating gluon fields, measuring gluon observables, calculating quark propagators, and calculating hadron observables from them.  The tutorial is addressed to lattice theorists interested in an introduction to the use of QDP/C, to other computational and computer scientists who are interested in seeing features of the QCD software environment that may be of value to other communities, and to those interested in seeing first-hand how lattice QCD works.  Tutorial participants will have guest accounts on the MIT Blue Gene/L.

Presenter: John Cary (TechX Corp)

The VORPAL computational framework has been used to compute the properties of a large number of physical systems, including those with plasma, fluids, electromagnetics, electrostatics, and with or without collisions, ionization, wall emission and other processes. VORPAL is highly flexible; it can execute in serial or parallel and in any dimensionality. It can resolve time scales explicitly or implicitly, making use of scalable solvers. As a parallel computational application, VORPAL has been shown to make effective use of thousands of processors. This tutorial will introduce VORPAL usage, starting from the simplest problems of electromagnetic propagation in vacuum, to the inclusion of charged particles, boundaries and other phenomena. Included will be the topics of problem definition, job submission, and data analysis.

Presenters: Phil Colella, Brian van Straalen, and Dan Martin (Lawrence Berkeley National Laboratory)

Problems in areas such as astrophysics, combustion, magnetic fusion, systems biology, and climate change are described by various generalizations of the classical elliptic, parabolic and hyperbolic partial differential equations of mathematical physics. The choices regarding mathematical models, discretization methods, software design and implementation all strongly interact with one another and with fundamental mathematical questions that underlie these choices. This places a premium on the availability of a diverse and agile software toolset that enables the use of computational experimentation to explore the design space. The Applied Partial Differential Equations Center for Enabling Technologies (APDEC), has developed a collection of algorithmic and software components that can be assembled to simulate a broad range of complex multicomponent physical systems in which partial differential equations play a central role. These tools emphasize three features: Capability - supporting the broadest range of applications and ability to combine the tools in as many ways as possible; Expressiveness - hiding non-essential details from users that don't need them, while providing access to them for users desiring to customize the software; and Performance - implementation to obtain the highest possible performance on high-end parallel platforms. In this tutorial, we will describe the APDEC software design, give examples of how to use these tools in several applications taken primarily from fluid dynamics, and discuss the future development plans for these tools.

Presenters: Erik Boman and Karen Devine (Sandia National Laboratories)

The largest scale scientific and engineering applications need to be executed on computers whose memory is physically distributed. For optimum execution time, an important consideration is how to best divide (partition) the data and work among processors to achieve load balance and also reduce communication. For dynamic or adaptive applications, the load balancing needs to be repeated periodically. We present the Zoltan software toolkit, a widely used freely-available software package for load balancing, supported by the CSCAPES Institute. Zoltan provides a set of useful abstractions, for both static and dynamic load balancing, allowing partitioning approaches, including recursive coordinate bisection, space-filling curves, graph partitioning and hypergraph partitioning. In this tutorial we show examples of how to use Zoltan for both static partitioning and dynamic load balancing, and discuss recent features such as the parallel hypergraph partitioner.

Presenters: Rob Falgout (Lawrence Livermore National Laboratory), Mike Heroux (Sandia National Laboratories) and David Keyes (Columbia University)

Multiscale, multirate scientific and engineering applications possess resolution requirements that are practically inexhaustible and demand execution on the highest-capability computers available. While the variety of applications is enormous, their needs for mathematical software infrastructure are surprisingly coincident; moreover the chief bottleneck is often the solver. At their current scalability limits, many applications spend a vast majority of their operations in solvers, due to solver algorithmic complexity that is superlinear in the problem size, whereas other phases scale linearly. Furthermore, the solver may be the phase of the simulation with the poorest parallel scalability, due to intrinsic global dependencies. This tutorial introduces some of the world’s most widely distributed, freely available, scalable solver software focused on relieving this bottleneck. Solver software directly supported under TOPS includes: hypre, PETSc, SUNDIALS, SuperLU, TAO, and Trilinos. Transparent access is also provided to other solver software through the TOPS interface.

Presenters: Sean Ahern (Oak Ridge National Laboratory) and Allen Sanderson (University of Utah)

Understanding scientific datasets generated at the DOE high-performance computing facilities is becoming increasingly difficult as dataset sizes and complexity grow beyond the scale that is approachable by traditional analysis techniques. VACET is delivering scalable solutions to real-world visualization and analysis problems through the use of the SCIRun and VisIt visualization systems. The SCIRun framework has long been a focus of research and development at the SCI Institute and has been the test bed for significant fundamental research in visualization techniques and their applications to scientific problems. VisIt is a turnkey application for data exploration, code assessment, and quantitative analysis suitable for use on petascale datasets. It has a scalable architecture for running data operations on a parallel machine to leverage resources "close to" the data and supports a client-server model for effective remote visualization use. This tutorial will cover usage of both tools, demonstrating the component architecture and extensibility of each.

Presenters: Rob Armstrong (Sandia National Laboratories), David E. Bernholdt and James A. Kohl (Oak Ridge National Laboratory) and Gary Kumfert (Lawrence Livermore National Laboratory)

Component-based approaches increase software developer productivity by helping to manage the complexity of large-scale applications, simplifying collaborative software development across distances and disciplines, and facilitating the reuse and interoperability of code. The Common Component Architecture (CCA) was designed specifically for high-performance scientific computing. It supports language-neutral component-based application development for both parallel and distributed computing without penalizing the underlying performance, and with a minimal cost to incorporate existing code into the component environment. The CCA environment is also well suited to the creation of domain-specific application frameworks, in which the rich domain-specific computational infrastructure is cast as components. We will cover the concepts of components, the CCA's particular approach to components, including the Babel language interoperability tool, and component-based scientific applications. The CCA tools and example software will also be available for download. CCA development is led by the SciDAC Center for Technology for Advanced Scientific Component Software (TASCS).

Members of the Open Science Grid (OSG) will show through a laptop and with posters how a site joins the OSG -- how the software is installed and the processes needed to join. They will also discuss the software installation and process for a new community (Virtual Organization) to use the distributed facility and the workflow and data movement services that are available . A demonstration of the LHC CMS experiment analysis on the OSG will be included as an example application.