Application Performance Tools
The Application Performance Tools group conducts research and development and manages vendor interactions in the development and use of software tools and middleware needed to assess and improve application performance on current and emerging HPC systems. The group is primarily focused on issues that arise when using extreme-scale systems and is actively working in the areas of high-performance communication, user- and operating system- level run-time middleware, compiler-based code transformations, file-systems, and performance characterization.
To fully utilize the capabilities of high-end computer systems, we provide the necessary understanding of technical hurdles and the means for surpassing these barriers. We work to ensure an efficient development environment for all aspects of complex scientific application programs.
Middleware -- Open MPI, Networking research,
InfiniBand, collective algorithms, vendor collaborations (Mellanox
- Compliers based
transformations -- Open64, research in complier feedback/transformations
tolerance -- Basic R&D in proactive fault tolerance at middleware
layers (MPI) and below (Colorado State Univ.)
-- Confidence, important performance measurements with characterize
system behavior at scale (latency, noise, etc.)
& operating systems -- Basic R&D in highly scalable runtime
systems, Scalable Tools Communications Infrastructure (STCI), Colony,
GPU aware computing, university collaborations (Univ. of Illinois at Urbana-Champaign)
Middleware -- Removing secondary storage barriers for extreme scale
systems, I/O Forwarding Scalability Layer(IOFSL)
debugging -- Allinea Distributed Debugging Tools (DDT), vendor
tracing -- VampirTracevendor collaboration (Dresden University of Technology)
| The Applications Performance Tools group provides both research and support to
benefit HPC users with these technologies.