Thursday 1 February 2018 - 14:00hrs GMT | 15:00hrs CET
HPC applications have a need for persistent storage for reading initial conditions, writing the final solution, and checkpointing. Checkpointing frequency increases with larger scale simulations due to the increased risk of node failure. This subsequently increases the load of the I/O subsystem and the Darshan profiling tool has been developed to help code developers understand their application’s I/O characteristics. The HPC workload characteristics can then be used to help code developers optimise the I/O of their applications and increase data throughput. The I/O characteristics can also help HPC system administrators configure parallel storage systems to better support the HPC user community.
In this 30 minute live webinar we will :
- Give an introduction to parallel storage/file systems;
- Discuss parallel I/O models available for MPI codes;
- Provide programming hints and tips on how to efficiently write I/O for MPI codes;
- Introduce the Darshan profiling tool, including a live demo.
About the Presenter
Wadud Miah is employed at NAG and is working on the Performance Optimisation and Productivity (POP) project. His background is in computational science, HPC, and parallel programming. He has worked in multiple roles in HPC, helping researchers improve their productivity. His other roles include code development, teaching computer programming and POP outreach.