John Linford

Numerical simulations of chemical kinetics are a critical component of aerospace research, Earth systems research, and energy research. These simulations enable a better understanding of the evolution of chemical species over time in domains as diverse as climate and weather prediction, combustion simulation, and air quality prediction. The time-to-solution in these simulations can be improved by over 30X via computational accelerators like Graphical Processing Units (GPUs) or the Intel Xeon… Show more

Numerical simulations of chemical kinetics are a critical component of aerospace research, Earth systems research, and energy research. These simulations enable a better understanding of the evolution of chemical species over time in domains as diverse as climate and weather prediction, combustion simulation, and air quality prediction. The time-to-solution in these simulations can be improved by over 30X via computational accelerators like Graphical Processing Units (GPUs) or the Intel Xeon Phi coprocessor, but the state-of-the-art tools for chemical kinetics do not support accelerators. ParaTools will develop a code generation tool called "Kppa" into a production-grade product that translates a high-level description of a chemical reaction network into simulation code that supports computational accelerators to significantly reduce time-to-solution. The generated code will provide the same software interface as existing tools to ensure immediate compatibility with popular codes like GEOS-Chem, WRF-Chem, MCM, etc. Kppa will include an online user productivity environment called "Kppa Cloud" for the development, testing, and benchmarking of chemical simulation codes. Kppa Cloud will enable users to graphically formulate new chemical reaction networks, maintain a library of chemical mechanisms, develop new mechanisms collaboratively, generate simulation code, explore the computational and numerical characteristics of the generated code, and test the generated code for stability and correctness. Kppa will enable supercomputer-level performance on smaller computers with lower costs, lower barriers to entry, and enable the rapid creation of high-performance kinetics simulations. Kppa will build on open source technologies to be backward compatible with the state-of-the-art in modeling and simulation and employ a modular design enabling extensibility to the computer architectures of the future. Show less

Many NASA programs, such as the Global Modeling and Assimilation Office
(GMAO), use atmospheric models to further the understanding of complex
Earth systems. NASA's microscale, mesoscale, and tropospheric simulations
of air quality, weather prediction, and climate change spend 60-90% of
their computational time in simulations of chemical kinetics.
KPPA provides a high-level framework for describing chemical
mechanisms and translating these into numerical simulations of… Show more

Many NASA programs, such as the Global Modeling and Assimilation Office
(GMAO), use atmospheric models to further the understanding of complex
Earth systems. NASA's microscale, mesoscale, and tropospheric simulations
of air quality, weather prediction, and climate change spend 60-90% of
their computational time in simulations of chemical kinetics.
KPPA provides a high-level framework for describing chemical
mechanisms and translating these into numerical simulations of the
atmosphere by generating highly-optimized code for traditional, multi-core,
and accelerated computer architectures. KPPA retains compatibility
with existing models while enabling larger, more descriptive, and
higher-quality simulations of atmospheric chemical kinetics on smaller,
cost-effective computers. By combining knowledge of sparsity in the model
data with parameterizations of the target hardware, KPPA will generate code
that will not only improve the performance of existing models executing on
traditional architectures, but will also allow it to target future parallel
optimization opportunities. NASA atmospheric modeling applications will
benefit from significantly reduced time-to-solution on supercomputing
resources without sacrificing solution accuracy. This will translate to
overall improvements in high-performance numeric simulations of the
atmosphere for use in climate, weather, and air quality applications. Show less

A complete understanding of a flexible helicopter rotor flight envelope
requires computational resources beyond the reach of most design engineers.
However, the advent of desktop computers with multi-core CPUs and GPUs makes such
computations affordable if the software is designed to use multi-core hardware efficiently.
Phase I demonstrated the feasibility of a re-engineering tool that facilitates software modernization.
In Phase II, ParaTools and Sukra Helitek jointly offer… Show more

A complete understanding of a flexible helicopter rotor flight envelope
requires computational resources beyond the reach of most design engineers.
However, the advent of desktop computers with multi-core CPUs and GPUs makes such
computations affordable if the software is designed to use multi-core hardware efficiently.
Phase I demonstrated the feasibility of a re-engineering tool that facilitates software modernization.
In Phase II, ParaTools and Sukra Helitek jointly offer to develop the software re-engineering tool prototyped in Phase I and to apply Phase I innovations to RotCFD, an Integrated Development Environment for rotors.
RotCFD is the second generation of the Rot3dc rotorcraft simulation tool developed and commercialized by Sukra Helitek and licensed and used by the Army, NASA, Navy, and the rotorcraft industry. Rot3dc has been successfully used in major rotorcraft initiatives including the V-22 (Osprey), RAH-66 (Comanche) and Sikorsky's Cypher UAV.
The tools and the technical expertise developed will be invaluable in modernizing and improving
the performance of application codes in addition to enhancing the current capabilities of RotCFD.
Such tools are important to maintaining the U.S. government's leadership
position in the helicopter world, making it possible to produce and validate
high quality designs at an affordable price. Show less

Principal investigator responsible for all technical design, reporting and documentation on the project. Coordinate all interactions between project partners to ensure that work proceeds according to contract agreements.

Create and deploy a complete development environment for porting and tuning parallel Linux applications on Microsoft Windows 64-bit with Microsoft MPI for use on Windows Azure or Windows-based clusters.

Develop and debug, implement new features, and improve usability. Also train, tutor, and support TAU users at high performance computing installations worldwide.

Numerical simulations of chemical kinetics are critical to addressing urgent issues in both the developed and developing world. Ongoing demand for higher resolution models with larger chemical mechanisms drives exponential growth in computational cost: many models spend over 90% of their runtime simulating chemical kinetics. Energy efficiency and renewable energy system research and development depend on simulations involving thousands of chemical species and reactions, but there are no… Show more

Numerical simulations of chemical kinetics are critical to addressing urgent issues in both the developed and developing world. Ongoing demand for higher resolution models with larger chemical mechanisms drives exponential growth in computational cost: many models spend over 90% of their runtime simulating chemical kinetics. Energy efficiency and renewable energy system research and development depend on simulations involving thousands of chemical species and reactions, but there are no general analysis tools that can handle mechanisms of this size. Simulations of more than a few hundred species or reactions are hand-tuned, ad-hoc solutions that will ultimately become obsolete. ParaTools will address this need by improving its “Kppa” general analysis tool for chemical kinetics to facilitate coupling with high resolution models and to support large chemical mechanisms. Phase I will explore the feasibility of methods for large mechanism support including flux analysis for sub-cell parallelization and mechanism reduction, dynamic mechanism selection based on environmental conditions, and iterative methods for large sparse systems. Phase I will also improve Kppa as a general analysis source code generator by implementing accelerated analysis methods that use many-core and multi-core devices and/or GPUs to reduce mechanism analysis, support for non-Arrhenius reaction rates, and an interface for coupling Kppa-generated code with high resolution models. Phase II will implement large mechanism support based on Phase I findings. Pre-coupled open source model packages containing Kppa-generated source coupled with a multi-physics or flow code will be provided in Phase I to facilitate commercialization through Phase II and beyond. The improved Kppa tool will reduce time-to-solution by combining the latest numerical and algorithmic developments with accelerated computing technology to enable supercomputer-level performance on smaller computers with lower costs. Show less

Using High Performance analysis tools created by ParaTools, Inc to assess the usefulness of these tools on a test code written in C++, OpenMP and MPI: GraphBLAS.

GraphBLAS is a linear algebra library for solving graph applications such as the common BFS (Breath First Search algorithm), most commonly used to find the shortest path between two nodes or vertices.

Development and benchmarking.

ThreadSpotter automatically analyzes application performance, rates performance problems, suggests fixes, and provides insights and statistics to quickly assess and resolve inefficient cache use. ParaTools ThreadSpotter expands these capabilities to support multi-language distributed memory applications and integrates with the TAU Performance System® to analyze data movement across the memory hierarchy on each compute node.

The Department of Energy and other federal agencies have made significant investments in high performance software engineering tools, yet these tools still lack advanced problem identification capabilities. At the moment, users must rely heavily on their own experience and intuition to interpret software performance data, identify the root cause of a software performance problem, and ultimately resolve the problem. This limits tool adoption by small companies and independent software vendors… Show more

The Department of Energy and other federal agencies have made significant investments in high performance software engineering tools, yet these tools still lack advanced problem identification capabilities. At the moment, users must rely heavily on their own experience and intuition to interpret software performance data, identify the root cause of a software performance problem, and ultimately resolve the problem. This limits tool adoption by small companies and independent software vendors, particularly in the manufacturing sector, who lack extensive in-house software engineering experience.
This project is developing a complete “production grade” software performance engineering product that lowers the barriers to entry for novice users and enhances their ability to mine actionable information from software performance data. The new product presents a simple, intuitive, and systematic user interface that guides users through performance engineering workflows and uses advanced cloud-hosted performance analysis services to offer unprecedented data analysis and problem identification and resolution capabilities. Show less