Workshop on Parallel Code Optimization
On Thursday, March 16, 2017, KAUST Supercomputing Lab(KSL) is organizing a workshop on Parallel Code Optimization in partnership with Seagate, Intel, and Cray.
In the morning session, Seagate, world leader in storage solutions, is presenting the best practices for parallel I/O. The ClusterStor Lustre parallel file system is designed to provide high performance by maximizing parallelism and avoiding bottlenecks. Understanding the file system tuning options and application I/O design can help in achieving the best results for different usage cases. This training will focus on the interaction of the file system and application, as well as issues in programming for parallel I/O. It will cover the architecture, general parallel I/O models, file system tuning, I/O tracing, MPI-IO usage, and general best practices.
The afternoon session will be dedicated to code optimization where KSL CS team, Seagate, along with Intel and Cray application engineers will help users to tune and optimize their codes for computation, scalability and I/O: bring your codes and problems.
We are looking forward to seeing you there.
Bilel Hadri, Workshop Chair
Rex Tanakit is leading the Performance and Systems Engineering group at Seagate Cloud Systems , a leader in the clustered storage and parallel file system. He interacts daily with researchers and scientists in the high performance computing (HPC) area. As CPU’s capability and speed increases, the performance bottleneck shifts to I/O - how fast data can be moved in and out of CPU. This is Rex’s current focus. Prior to joining Seagate, Rex was a Vice President of Engineering at Panasas for nearly 10 years and Director of Engineering at SGI for 12 years. He was responsible for Performance Engineering and benchmarking. Rex received his BS in Chemical Engineering from UC San Diego and MS from UCLA.
Bill Loewe has contributed for more than 15 years in various roles in HPC applications performance and tuning within industry at Seagate Technologies, Panasas, and government agencies at the US Department of Energy, Lawrence Livermore National Laboratory. Dr. Loewe has provided talks and co-authored several papers for technical conferences and HPC research journals, and his education includes a Ph.D. from University of California, Santa Barbara, CA and a M.Sc. from the California State University, Hayward, CA, USA
- 8am Registration
- 8.15am Welcome
- 8.20am to 11.30am Parallel I/O Performance background
- Overview of Parallel I/O and Seagate Clusterstor Lustre
- Optimal configuration for KAUST
- Recommended tuning options for HPC workloads
- Tools to identify issues
- Visibility/monitoring on servers/clients
- File/directory striping
- I/O tracing
- POSIX and MPI-IO
- Use of POSIX and directly accessing file system
- Use of MPI-IO as middleware I/O library, including MPI-IO hints
- Best Practices: How applications should use Lustre
- DOs and DONTs
- Access patterns
- Metadata Usage
- Afternoon session at 1.30pm: Parallel code optimization with KSL scientists, Seagate, Intel and Cray engineers
- Shaheen Applications Hands-on session
To register to the event, please click here
All lectures will take place in the Lecture hall 2 room 2322 in building 9 (black square).
The workshop location is a 10 min walk from the KAUST Inn (red square) and the campus diner (blue square) for the regular lunch is 5 min away from the workshop location.
The slides are available here.
The event will be followed up by an afternoon session dedicated to developers and users of Shaheen by KSL scientists in collaboration with Seagate, Intel and Cray engineers to tune your applications.
Please contact us at email@example.com if you need further information.