Scientific Software Days

Speakers and Abstracts for Scientific Software Days 2012

My Monitor is Bigger than yours: Running Visualization with DisplayCluster on Tiled Arrays

Greg Abram and Greg Johnson
Texas Advanced Computing Center

DisplayCluster is a software environment for interactively driving large-scale tiled displays. The software allows users to interactively view media such as high-resolution imagery and video, as well as stream content from remote sources such as laptops / desktops or high-performance remote visualization machines. Many users can simultaneously interact with DisplayCluster with devices such as joysticks or touch-enabled devices such as the iPhone / iPad / iTouch or Android devices. Additionally, a Python scripting interface is provided to automate interaction with DisplayCluster.

Biophysics and HPC: The UltraScan XSEDE Science Gateway

Borries Demeler
UT Health Science Center San Antonio

UltraScan is a software package for the analysis of biophysical data from experiments investigating solution properties of macromolecules. This presentation will discuss the analysis workflow used to solve a complex computational problem. We describe the UltraScan-3 XSEDE Gateway, which provides a unique analysis environment that combines multiple tools and layers, including MySQL databases, the XSEDE HPC infrastructure, multi-platform GUI software, a PHP-based LIMS web frontend, and the Apache Airavata Generic Factory (GFAC) into a powerful analysis toolchain. Together, these components provide an easy-to-learn and user-friendly, yet very flexible analysis environment for remote HPC analysis and large data management.

Having It Both Ways: Eclipse PTP on Desktop and Cluster

Doug James
Texas Advanced Computing Center

This is an overview of Eclipse PTP, a robust, open source, multi-platform Integrated Development Environment (IDE) supporting the development of parallel applications in C, C++, and Fortran. PTP also delivers customized, tight integration with XSEDE (including TACC's) clusters. Developers can use PTP on their desktops to develop, build, debug, manage, submit and monitor remote jobs on XSEDE resources; they can also create "synchronized projects" that reside on both their desktops and remote clusters. Eclipse PTP offers a promising path for improving both code quality and developer productivity; the presenter hopes to increase its visibility and promote its use.

Bring on the Stampede: Coding with the Xeon Phi

Lars Koesterke
Texas Advanced Computing Center

In January of 2013, TACC will deploy Stampede, the first large-scale cluster deployment to include Intel Xeon Phi (MIC) co-processors. The Xeon Phi is a highly parallel architecture combining many x86 cores into a single chip. Leveraging the standard Intel Xeon technology, developers are able to program the Xeon Phi using standard threading models. This highly-parallel processor architecture with a portable threading model promises substantial performance gains for highly-parallel computing workloads. In this presentation, we discuss porting computationally intensive algorithms, scaling to large numbers of threads and cores, and using Stampede and the Intel Xeon Phi.

Reading/Writing Large Parallel Files without upsetting Supercomputer Administrators

Robert McLay
Texas Advanced Computing Center

Large parallel filesystem are magical. They are a collection of computers and disks that act as if they are one very large disk. However the magic only goes so far and they are somewhat fragile. This talk will explain what issues are and how to avoid the pitfalls. As an example this talk will show that it is possible to write single large files efficiently in parallel.

Fortran speed in Python: Numba the Python LLVM Compiler

Travis Oliphant
Continuum Analytics, Inc

Numba and NumbaPro is a Python bytecode to LLVM translator that allows creation of fast, machine code from Python functions. The Low Level Virtual Machine (LLVM) project is rapidly becoming a hardware-industry standard for the intermediate representation (IR) of compiled codes. Numba's high-level translator to the LLVM IR provides Python the ability to take advantage of the machine code generated by the hardware manufacturers contributions to LLVM. Numba translates a Python function comprised of a subset of Python syntax to machine code using simple type inference and the creation of multiple machine-code versions. In this talk, I will describe the design of Numba, illustrate its applications to multiple domains and discuss the enhancements to NumPy and SciPy that can benefit from this tool.

Everything about scientific software documentation that wasn't in the manual

Aleksandra Pawlik
Open University

How does documentation fit into different contexts of scientific software development? The answer is different for end-user developers and in the case of software used by a wider scientific community. The talk will examine these scenarios and will look in detail at the SciPy Documentation Project, to illustrate how documentation can be crowdsourced in scientific software development. This will lead us to consider how crowdsourcing documentation relates to open source software development and community building.

Data Processing for the NASA GRACE Mission

Peter Nagel and Steve Poole
Center for Space Research, UT Austin

The NASA GRACE satellite mission obtains a set of global high-resolution gravity measurements which are used to solve for an Earth gravity model every month. The procedures used to create this monthly gravity model will be discussed, including the challenges involved in getting large amounts of data ready for processing, keeping large data files on-line during the processing, condensing the data into manageable-size files for use in creating multi-year mean gravity models, and archiving the results. The software used to solve for the gravity model, the Advanced Equation Solver for Parallel Systems (Aesop), will be discussed.

Scaling the Future: How to Practice Open Science

Will Schroeder
Kitware, Inc.

The growing complexity and scale of research demands that we embrace new methods if we are to support future scientific progress. These methods must be based on the three pillars of Open Science: Open Access, Open Data, and Open Source. This presentation describes some practical means to practice open science, with examples from analytics, computational chemistry, big data, medical computing, video analysis, and visualization.

Sharing is Caring: NumFOCUS and Open Scientific Code Initiative

Andy R. Terrel
Texas Advanced Computing Center

Scientific software is complex, hard to install, and usually written by overworked indentured servants, i.e. graduate students. Perhaps more importantly, it is the large basis of the modern scientific method, but unlike proofs and lab methodologies, it is often hidden from our peers. To address these issues, a number of initiatives to promote open scientific codes have emerged from various organizations. I will focus on the NumFOCUS foundation, a new organization promoting the use of accessible and reproducible computing in science and technology. I will also survey a number of other resources available for scientists to share their code and experiences.

Filesystems: Addressing the Last-mile "problem" in Services-Oriented/Cloud Computing

George K. Thiruvathukal
Loyola University Chicago and IEEE/AIP Computing in Science and Engineering

We have designed and implemented RestFS, a software framework that provides a uniform, configurable connector layer for mapping remote web-based resources to local filesystem-based resources, recognizing the similarity between these two types of resources. Such mappings enable programmatic access to a resource, as well as composition of two or more resources, through the local operating system’s standard filesystem application programming interface (API), script-able file-based command-line utilities, and interprocess communication (IPC) mechanisms. The framework supports automatic and manual authentication.

As part of this talk, we demonstrate a new filesystem that interfaces to the SketchFab.com service to support 3D model visualization. For more information about the NOFS project, visit http://nofs.googlecode.com, where you can also download and use all of our reference filesystems.

Tex-MECS: Infrastructure for managing ensembles of computations

Michael Tobis
University of Texas Institute for Geophysics

The yt Project: Growing and Engaging a Community of Practice

Matthew Turk
Columbia University

In this talk, I will present on the yt Project an analysis and visualization tool for astrophysical simulations. I will provide a brief motivation for its development, describe its capabilities for analysis of astrophysical simulation data, and then describe the method by which the community attempts to balance pragmatic, goal-driven development with best practices in software development and high-performance computing.