-
Notifications
You must be signed in to change notification settings - Fork 0
Sync meeting 2023 11 28 with CernVM FS developers on Best Practices for CernVM FS on HPC tutorial
Kenneth Hoste edited this page Dec 8, 2023
·
1 revision
- https://github.com/multixscale/cvmfs-tutorial-hpc-best-practices
- online tutorial, focused on (Euro)HPC system administrators
- aiming for Fall 2023 (Sept-Oct-Nov)
- collaboration between MultiXscale/EESSI partners and CernVM-FS developers
- tutorial + improvements to CernVM-FS docs
- similar approach to introductory tutorial by Kenneth & Bob in 2021, see https://cvmfs-contrib.github.io/cvmfs-tutorial-2021/
- format: tutorial website (+ CVMFS docs) + accompanying slide deck
- EESSI/MultiXscale: Kenneth, Lara, Alan, Bob, Thomas
- CVMFS: Valentin, Laura, Jakob
- final tally on registrations: exactly 200 people!
- tutorial website: https://multixscale.github.io/cvmfs-tutorial-hpc-best-practices
- "preview" website (boegel.github.io/...) is now 404
- last chance to give feedback on contents
- Kenneth needs to take into account feedback shared by Laura in Mattermost (WIP)
- now in "only important things can still be changed" mode
- rien-ne-va-plus w.r.t. contents at 12:00 CET
- slide deck to drive tutorial (WIP)
- practical info to share with registered attendees
- Zoom link (final)
- tutorial website
YouTube live stream- dedicated channel in EESSI Slack:
#cvmfs-best-practices-hpc
- should we also mention CVMFS Mattermost? => not in email
- slide to let
JakobValentin say a word- while showing slide 3
- agenda - timing
(*): incl. hands-on demo [13:35-14:00 CET] Introduction to CernVM-FS [14:00-14:20 CET] * EESSI [14:20-15:00 CET] * Accessing repositories [15:00-15:30 CET] (coffee break) [15:30-15:50 CET] * Configuring CernVM-FS on HPC infrastructure [15:50-16:10 CET] * Troubleshooting [16:10-16:30 CET] * Performance aspects [16:30-16:40 CET] * Containers [16:40-16:50 CET] Creating a CernVM-FS repository (birds-eye view) [16:50-17:00 CET] Q&A
- live demo of
- structure of EESSI repo + using EESSI
- installing & configuring CVMFS client + proxy + Stratum 1
- show performance impact of no proxy + distant Stratum 1
- start TensorFlow container via unpacked.cern.ch
- troubleshooting (firewall problem, incorrect ACL in Squid proxy config)
- ISC'24 tutorial submission
- deadline Fri 8 Dec'23
- add Valentin + Jakob + Laura as co-author
- join Zoom by 13:00 CET to prepare
- Valentin, Kenneth, Alan
- CernVM-FS: Laura, Valentin, Jakob
- EESSI/MultiXscale: Bob, Kenneth, Lara
- Practical
- T minus 6 days...
- 152 people have registered so far...
- Should we send out another reminder?
- regular Zoom session
- via CERN => Valentin
- can also use Zoom setup at Univ. of Oslo (via Terje)
- no need for webinar mode, we can make sure that participants can not unmute on join
- cloud recording
- with support for streaming to YouTube (backup recording)
- dedicated Slack channel in EESSI Slack
- send out practical info on Mon 4 Dec around 09:00 CET
- send out message to notify people that practical info will be sent on Monday 4 Dec
- last minute sync Mon 4 Dec at 10:00 CET
- agenda
- 13:30 - 17:00 CET
- 13:30 CernVM-FS
- [14:00 - 14:15] EESSI
- [14:15 - 14:45] Access
- client setup
-
sudo cvmfs_config setup
- required to create
cvmfs
user configure autofs - can skip autofs configuration
- required to create
- client config
- warning on using direct proxy
- fstab based mounting instead of autofs, or manual mount
-
cvmfs_config chksetup
after custom client config - show hands-on
-
- proxy server
- stateless
- show hands-on
- recommendations
- at least two (maintenance reason)
- rule of thumb: 1 powerful proxy per 100-500 nodes
- 10Gbit link to nodes
- SSD storage
- decent CPU
- depends on workload mix
- very easy to scale up, especially via round-robin DNS setup
- replica server
- required resources, monitoring
- pre-run snapshot because this takes time
- can use S3-like as backend storage (CEPH, Azure blob, AWS S3, ...)
- alternative ways
- cvmfs-exec
- cvmfs in container with
apptainer --fusemount
- alien cache
- client setup
- [15:00 - 15:15] coffee break
- [15:15] Configuration for HPC => Bob
- diskless
- prefered => loopback cache on shared FS
- client cache in memory
- stealing memory
- not extensively tested, complex to support this in CernVM-FS
- alien cache on shared FS
- eaiest to configure
- may overload shared FS
- offline
- prefered => proxy and/or replica server in local network
- preload alien cache
- drop security bit?
- export CVMFS to other FS
- sync subdirs of CMVFS repo to filesystem like NFS
- same problems as installing software on NFS filesystem
- heavy-weight process
- needs to be kept in sync
- https://cvmfs.readthedocs.io/en/stable/cpt-shrinkwrap.html
- NFS export is not recommended
- Parrot not recommended anymore
- diskless
- status update tutorial contents
- https://multixscale.github.io/cvmfs-tutorial-hpc-best-practices
- CernVM-FS section: done
- EESSI section: done
- in progress (see preview at https://boegel.github.io/cvmfs-tutorial-hpc-best-practices)
- Access section: client install + config, proxy, (private) Stratum-1
- will finish this today
- Troubleshooting section
- incorrect "
CVMFS_REPOSITORIES can be used to limit access to specific repositories
"
- incorrect "
- Performance section
- MPI startup: impact of proxy
- drop OS jitter, drop CDN
- Storage backends => DROP?
- HPC section
- Containers section
- Valentin can look into this section
- incl. example to demo
- Creating CVMFS repo section => very short, refer to 2021 tutorial
- Access section: client install + config, proxy, (private) Stratum-1