Speaker: John Linford, ParaTools
Date: Thursday, May 25, 2017
Location: NASA/LaRC, Bldg 2102, Room 202
Host: Eric Nielsen, NASA/LaRC, Dana Hammond, NASA/LaRC
The TAU Performance System® is a powerful and highly versatile profiling and tracing tool ecosystem for performance engineering. Developed over the last two decades, TAU has evolved with each new generation of high performance computing (HPC) systems and presently scales efficiently to hundreds of thousands of cores on the largest machines in the world.
TAU can be applied in a portable way to codes written in Fortran, C, C++, Java, and Python, which utilize MPI message communication and multi-threading (e.g. pthread, OpenMP) for execution across different parallel machines. TAU also supports cutting edge architectures like Intel Xeon Phi (MIC), GPU accelerators, and PGAS languages like UPC and SHMEM. TAU’s analysis tools include a parallel profile analyzer, a performance data-mining tool, and a performance database. Multi-language debugging, memory leak detection, and kernel-level performance analysis are extended capabilities found in the toolset.
This presentation will present TAU and demonstrate TAU usage with TAU Commander, a performance engineering workflow manager from ParaTools that dramatically improves the usability of TAU. TAU Commander lowers the barrier to entry for novice TAU users by presenting a simple, intuitive, and systematic user interface that guides users through performance engineering workflows and offers constructive feedback in case of error. We will demonstrate profiling, tracing, and data analysis of MPI, OpenMP, and hybrid parallel codes on traditional multi-core CPUs, many-core CPUS (i..e. KNL), GPUs, and distributed memory clusters.
Dr. Linford will be available for one-on-one consulting throughout the afternoon of May 25th.
Dr. John Linford’s research interests include emerging computer architectures, automatic code generation, performance analysis, and numerical simulation. He develops software tools for chemical kinetic simulation, computational fluid dynamics, software performance analysis, and software environment management. John works for ParaTools, Inc. in Baltimore, MD.