About
We present our early experiences with the Distributed Asynchronous Object Storage (DAOS) focusing on its usage for checkpointing with the real-world scientific application PALM. The presentation includes an introduction of the of large eddy simulation and its legacy Fortran IO checkpointing mechanism. A thin software layer is introduced to remove the hard-coded IO operations by generic API calls. This abstraction enables the application to support both Fortran Stream IO, MPI IO, as well as netCDF on top of HDF5 which is the principal foundation for using DAOS in our use-case. We present early performance numbers on a Cascade Lake test system with Optane DCPMM and a comparison with the Lustre-based infrastructure of the HLRN-IV phase 1 system. The talk also presents some of the lessons learned from adjusting the application's legacy checkpointing code to a modern software environment..
0:00:00 Slide 1
0:00:08 Slide 2
0:00:25 Slide 3
0:01:02 Slide 4
0:02:09 Slide 5
0:02:51 Slide 6
0:03:01 Slide 7
0:03:23 Slide 8
0:04:17 Slide 9
0:05:27 Slide 10
0:05:38 Slide 11
The talk
Credits and licence
- Institution
- CERN
- Department
- other events or meetings
- Speaker
- Steffen Christgau
- Year
- 2019
- Language
- English
- Source
- CERN
- Licence
- CERN, as CERN states it. CERN's terms
- Attribution
- © CERN. CERN provides its audiovisual media free of charge for educational and informational use, credited to CERN; this lecture is shown in CERN's own player.
- Who can watch
- CERN gives no rating. Invisible Network shows it to Teens profiles and older, judged from what the source says about it. How courses are placed