Choose timezone
Your profile timezone:
HEPiX Virtualisation Working Group Meeting
→
Europe/Zurich
CERN
CERN
Description
EVO meeting in the WLCG community:
You have BOOKED a meeting in EVO (http://evo.caltech.edu).
<xmp>
Title: HEPiX vwg
Description:
Community: WLCG
Password: lip
Phone Bridge
ID 2059756
pwd 1922
</xmp>
Meeting Access Information:
- Meeting URL http://evo.caltech.edu/evoNext/koala.jnlp?meeting=M9MIMi2I2BDaDM9a9nDD99
HEPiX Virtualisation WG meeting Minutes 2010-06-28
Present (in no particular order)
Tony Cass, CERN
Romain Wartel, CERN
Sebastien Goasguen, CERN,
Ulrich Swicherath, CERN
Martin Bly, RAL
Ian Collier, RAL
James Adams, RAL
John Gordon, RAL
Thomas Finnern, DESY
Michel Jouvin, GRIF
Abdeslem Djaoui, RAL
Thomas Finnern, DESY
Michel Jouvin, GRIF
Abdeslem Djaoui, RAL
Matthieu Puel
Keith Chadwick, FermiLab
Doug Olson, LBL
1. Comment on the previous meeting minutes
No comments
2. Virtualisation workshop follow-up
Tony reported on last weeks Multi-core and Virtualisation workshop. (Report attached to agenda.)
Agreement that CernVM-FS filesystem is in principle acceptablet as a mechanism for accessing experiment software from virtual machines, pending a security review.
Also need to check how it fits with the image creation policy - which currently assumes the experiment software is included in the image.
3. Contextualisation Proposals
There was an extended discussion about issues around contextualisation.
No firm names proposed beyond that one would be a prolog and one an epilog and would likely contain those words in their respective names. Ulrich, Romain and Sebastien volunteered to test mechanisms for contextualisation with a prolog and epilog scripts.
Lifetimes of machines using pilot job frameworks
Ulrich then raised that we need a way to manage pilot job frameworks differently from regular jobs. He described how lxcloud at the moment allows new VMs to accept jobs for 24 hours, then they stop accepting new jobs, drain and when all work has finished shutdown and are effectively destroyed. Since no VM lives for more than a few days at most it is natural and easy to ensure that images are patched and up to date.
For pilot jobs it is more complicated as they typically need to run for longer, many are small and we do not want to instantiate VMs for each one etc.
They could have a, site configurable, fixed lifetime - one week say - but then in order to be efficient, one would have to be able to work out if there is enough lifetime left to run another job. This would likely result in inefficiency.
Alternatively, dynamic lifetimes could be facilitated with some mechanism (perhaps a standard named file in the VM filesystem) for communicating a) from the VM that all useful work is done and the site can now destroy the VM, or b) from the site that there is a critical update required and that no new work should start and that theVM should shut down as soon as possible).
We need to either make a clear decision that we prefer one model, or alternatively find a way to support both. Further discussion is needed.
4. VM Image format
Limited discussion took place since Andrea was not available to present his document.
Romain observed that there is a clear choice between agreeing a single format or having multiple formats along with the administrative cost of converting between them. Romain also observed that without a standard format we need a recipe for how to instantiate each image.
There was some discussion of image inspection. Those taking part agreed that it while it has to be possible to inspect the contents of images, sites are unlikely to need to inspect all images and there is no need to support automated inspection.
More discussion needed on the image formats.
5. VMIC Plan
Owen not available so no further discussion.
6. Actions
- Tony to discuss with Dave Kelsey the implications for the image creation policy of using CernVM-FS.
- Sebastien to discuss image creation with Owen
- Tony to discuss VMIC planning and timetable
- Tony to discuss multiple hypervisor support with Andrea
- Ulrich, Romain and Sebastien volunteered to test mechanisms for contextualisation with a prolog and epilog scripts.
7. Next Meeting
Monday 7th July at 14:00 UTC
Tony to determine and post agenda.
There are minutes attached to this event.
Show them.