Choose timezone
Your profile timezone:
HEPiX Virtualisation Working Group "f2f" video conference
→
Europe/Zurich
CERN
CERN
Description
EVO meeting in the WLCG community:
You have BOOKED a meeting in EVO (http://evo.caltech.edu).
Title: HEPiX vwg
Description:
Community: WLCG
Password: lip
Meeting Access Information:
- Meeting URL
http://evo.caltech.edu/evoNext/koala.jnlp?meeting=vsvivIenenItIIauaIItas
HEPiX Virtualisation Working Group 'virtual' Face to Face 2010-04-22
Present (in no particular order):
Tony Cass, CERN
Romain Wartel, CERN
Ulrich Schwickerath, CERN
Martin Bly, RAL
David Kelsey, RAL
Keith Chadwick, FNAL
John Gordon, RAL
Ian Colier, RAL
Sander Klous, NIKEF
Thomas Finnern, DESY
Owen Synge, DESY
Hermann Hessling, DESY
Jie Tao, KIT
Ian Gable, University of Victoria
Michel Jouvin, GRIF
Andrea Chierici, INFN Bologna
Doug Olson,
Mattieu Puel, IN2P3
Steve McDonald, TRIUMF
Ewan MacMahon, Oxford
1. No comments on the previous meeting minutes.
2. Discussion of types of models this working group is addressing.
There was a prolonged discussion of if we should be addressing the model of virtual worker nodes that act in a very similar way to grid nodes of today, or if we address a more cloud like model. Ewan commented that, in his opinion, it was actually more complicated to contextualize worker nodes and integrate them fully with the site infrastructure then to operate in a cloud-like way. Ewan also commented that he thinks that pursuing a virtual worker node model negates many of the advantages of virtualization. Ian Gable commented that he viewed the virtual worker node model as a way to get the foot in the door for virtualization, and that clouds can come later. Tony commented that there has been opposition from sites to running in a completely cloud like way, and that the consensus has evolved that we can get acceptance for the virtual worker node model. Tony further elaborated that the virtual worker node model is simply a different way of running authorized VO work, exactly the same as if the payload would have come from the workload system, hence there should be no need to place more restrictions on virtual worker nodes than real ones. It was agreed that discussions with the VOs, especially the LHC VOs, is required to finalise the models and plans.
3. Review of "Policy on the Generation of Trusted Virtual Machine Images for Use on the Grid" Version 1.2
Document at: http://www.jspg.org/wiki/Policy_Trusted_Virtual_Machines
Dave went over the document point by point and people commented.
Definitions and scope:
- Ewan suggested that "VM combined image" could be "VM complete image".
- Several people suggested that the definition of "Tag" should be a UUID.
Policy Requirements:
- Dave to create version 1.3 based on the comments during the meeting.
- A Use case section explaining the working model should be added under point 1.
- Tony will clarify the situation with the VOs
Discussion Point:
- There is no need to assign liability to the endorser
- Points 2-5 to be reviewed tomorrow based on discussions of owen/ewan/martin doc.
- The policy shouldn't document best practice; there should perhaps be a pointer to a best practice document elsewhere, but it was not obvious that such a document exists.
4. Working with multiple hypervisors
- Andre installed KVM host machine another machine installed dom0.
- Andrea is able to run the same image on both on KVM and Xen
- Andrea is quite confident it's going to work, and expects it to be accomplished rather soon.
- Owen asked if the image works with Xen para-virt, and yes it does by including two kernels.
- Owen asked if Andrea had addressed the TLS glibc issue (xen para-virt). Andrea had not, however people were unsure if this is still an issue with SL5.
- Tony asked Andrea to produce a document describing his work and he agreed.
5. Image Cataloguing and Distribution
Martin went over the Virtual Machine Image Catalogue document created by Owen, Ewan, and Martin.
- Romain commented that they that the verbal description was a little complicated, but he didn't have the same feeling about the written document.
- Ewan commented that he thought that what was described was the simplest thing that would work given the requirements. This was seconded by Martin and Owen.
- Dave commented that he would like to see a meaningful Globally Unique identifier i.e. something that means something to humans.
Day 2: 2010-04-23
1. No opposition voiced to the previous minutes
2. Image Cataloguing and Distribution
- Owen commented that image distribution sub group had not fixed upon a push or pull model. He felt that Romain had presented convincing arguments for pulling.
- Ewan discussed how he felt a push model could work.
- Martin went over the VMIC architecture diagram.
- Sander expressed concern about images possibly being downloaded before their are filtered against the local site preferences. Ewan/Martin/Owen explained that it would be possible to filter against meta data, however it is of course necessary to download the image before it's hash can be checked.
- There was discussion that if being present in the local catalogue filesystem was enough to considered runable. Ewan/Owen/Ulrich pointed out that you still to programatically check the catalogue just prior to booting to ensure the information you have is up to date.
- Tony concluded the discussion by asking that an API specification for the VMIC be produced by Owen, Martin, Ewan, and Ulrich
3. Use cases & Image generation workflow
- Section 3 point 3 should be revised to include a pointer to currently accepted image metadata.
- Dave had no more questions for the group and will attempt to provide a Version 1.3 sometime next week.
Discussion then turned to how someone gets on the endorsers list:
- Dave favored nominating people to select Endorsers, for example VO Representatives, or Tony Cass to start.
- Ian Gable felt that the list of endorsers should not be exclusive but rather the list of people sites except should be exclusive.
- Ewan commented that VO id cards could be a useful mechanism for formulating a list of accepted endorsers.
- Dave emphasized that the job of being an endorser was rather demanding, and he would prefer that the list be small or at least small to start.
- Tony "stuck his neck out" to say that, in the end, the job of selecting appropriate endorsers would fall to the VOs.
- We concluded with starting with a small number of endorsers.
4. Next meetings
- April 30, 16:00 UTC, by EVO
- May 11, fate to face meeting coincident with GDB meeting, details to be sorted by John and Tony.
There are minutes attached to this event.
Show them.