-
Notifications
You must be signed in to change notification settings - Fork 0
meeting Apr 1 2021
Kenneth Hoste edited this page Apr 1, 2021
·
6 revisions
- date & time: Thu Apr 1st 2021 - 2pm CEST (12:00 UTC)
- (every first Thursday of the month)
- venue: (online, see mail for meeting link, or ask in Slack)
- agenda:
- Quick introduction by new people
- EESSI-related meetings in last month
- Application for CZI grant “Essential Open Source Software for Science”
- Progress update per EESSI layer
- Pilot repository: changes & status
- Usage of AWS resources
- Discussion with NVIDIA w.r.t. CUDA
- S4 NeIC project proposal + NESSI test lab
- Next steps
- Past & upcoming events
- Q&A
(by Bob, Kenneth)
- New people on the call:
- Jure Pečar from EMBL
- experienced EasyBuild user, helping out with BLIS evaluation
- Jure Pečar from EMBL
- ephemeral publish container is in Program of Work
- discussed inclusion of EESSI in default cvmfs-config repo
- Proposal written/submitted by Alan in collaboration with the UMCG (University Medical Center Groningen), focussing on rare diseases and supporting biomedical workflows with EESSI
- question by Victor: how will this impact EESSI in general?
- Alan/Kenneth: general goal should not be impacted, it's mostly focused on a particular use case, which probably means we should provide a bunch more bioinformatics software in our repo to support typical workflows used in rare disease studies
- planning to create new master key for Stratum-0
- only store on Yubikeys (+ physical backup like USB stick in safe storage)
- then get EESSI configuration into cvmfs-config-default repository, so any client can get easy access to EESSI
- Caspar: this would be good time to document native installation of CernVM-FS for EESSI
- by Jörg with help from Bob & Terje
- up and running in AWS instance (tx.large) in eu.west region (?)
- not included yet in latest EESSI configuration (0.3.0), but will be soon
- "it was fairly easy"
- currently using XFS, which was not recommended in older CVMFS versions -> to be checked with CVMFS developers
- question by Jure if he should set up a Stratum 1 at EMBL
- makes sense, eventually every "big" HPC site would want their own Stratum 1 anyway (to have a full copy of the repo, to protect themselves from network issues)
- how many Stratum 1 servers do we want and how should they be distributed?
- CVMFS devs warned us that we shouldn't have too many either...
- but not all Stratum 1 servers need to be included in EESSI configuration
- still looking for volunteers to go through the process to set up an additional Stratum 1 (in different AWS region)
- we should reach out to the Gentoo developers to check if/how we can help to get the problems that we run into (Lmod, bootstrap on POWER) resolved
- can we help with testing stuff, showing that it works, etc.?
- check if we can use ReFrame instead of pytest for the compatibility layer validation
- ask Victor if these compat layer tests can be run through ReFrame?
- the installation of 2021.02 was broken and has been removed from the repository
- ppc64le installation on hold for 2021.03, hoping that the upstream fix for the bootstrap script will get done soon
- some experiments with speeding up the install script by running multiple EasyBuild sessions in parallel, to overlap installations with different dependencies
- not fully working yet
- better approach would be to farm out installations to a Slurm cluster (via CitC in AWS)
- ppc64le on hold, until the bootstrap issue has been fixed upstream
- More or less the same software and hardware targets
- GPU installations on hold until we get green light from NVIDIA to include CUDA in the repository
- by Jörg
- the build script worked very well: the colors are very useful, the comments are nice, and it does some nice checks (e.g. if you are running it in a Prefix environment)
- we need to write some documentation for doing the software installations
- Jörg volunteers to do this by making a PR in the
docs
repository
- Jörg volunteers to do this by making a PR in the
- question by Jörg: would it help to invite more vendors?
- there's already quite a few very big companies joining the call, and we shouldn't make the group too large.
- ...
- big deployment of EESSI on several systems with different hardware, including GPUs
- limited access to a small group of users using permissions on
/cvmfs
- AWS credits are being spent on several things: Stratum 1, build nodes, test machines, etc
- Discussions with Azure are ongoing
- We need Packer-built images for Openstack as well
- Terje shows a demo of the infrastructure code / scripts for deploying dynamic infrastructure
- create/remove nodes
- grant access to users based on GitHub handles
- support for different node types
- available on the AWS login node
- interest in setting up a separate meeting to discuss how software stack on new clusters can be set up with a later transition to EESSI in mind?
- Bob will set up a doodle for this