Choose timezone
Your profile timezone:
HEPiX Virtualisation Working Group Meeting
→
Europe/Zurich
Phone conference
Phone conference
Description
<sl>
Dial +41227676000 and enter access code 0186715 followed by #
Key ##1 to mute/unmute your phone
</sl>
HEPiX Virtulisation WG meeting minutes 2010-03-15
Present (in no particular order):
Tony Cass, CERN
Ulrich Schwickerath, CERN
Belmiro Rodrigues Moreira, CERN
Romain Wartel, CERN
James Adams, RAL
Ian Collier, RAL
Martin Bly, RAL
Ewan MacMahon, Oxford
Jie Tao, KIT
Keith Chadwick, FNAL
Sonder Klous, NIKEF
Thomas Finnern, DESY
Owen Synge, DESY
Andrea Chierici, INFN Bologna
Mattieu Puel, CCIN2P3
Philippe Olivero, CCIN2P3
Ian Gable, University of Victoria
1. Amendment to minutes of 2010-02-15
- There is no confirmed time for the face to face meeting at HEPiX. It was not concluded as Friday as stated in the previous minutes.
- Tony commented a conclusion in the previous meeting was that there was consensus that data access is not something that needs to be contextualized.
2. Xen & kvm support
This was quite a lengthy discussion. I have attempted to make a reasonable summary and tease out some of the conclusions. Please feel free to correct me.
- After discussion there was a conclusion that KVM, Xen and Xen para-virtulized will be supported. If we there is someone with the expertise willing to do the work we could consider other hypervisors.
- Andrea now has a two machine testbed, he as already begun testing with supporting Xen and KVM and is confident it is simple with the .
- Andre asked if we should focus in particular on a particular Distribution, and if we need to consider SL4 (which is more difficult to use with both KVM and Xen).
- Ulrich responded that for CERN we can forget SL4.
- Tony commented that there is a clear reason not to bother with the SL4 with regard to the increased difficulty.
- Tony commented that if people what to create images other then SL5 then the onus is on them to make it work with multiple hypervisors.
- Keith commented that at FNAL they have demonstrated to their satisfaction that they can reconfigure an image to work with Xen or KVM.
- Ian Gable mentioned the pygrub tool useful for booting para-virtulized Xen images.
- Owen mentioned kpartx a useful for dealing with internal vm partitions.
- Ian Gable asked if we should define an image format i.e. a disk image containing LVM partitions, or a single ext3 partition. Andrea commented that he didn't think this fell under his sub task, but that his preference was for single partition.
- Andre asked the group if he should (1) begin by defining a list of supported virtual hardware layers or (2) if he should begin with a simple SL5 example and then expand it.
- Andrea would prefer the second approach, and the group supported this.
- Ewan commented that he hopes that virtual hardware auto detection will make the task of supporting multiple hypervisors.
- Keith commented that the management layer may end of configuring images to run on multiple hypervisors
- Owen commented that we should have the fallback position of providing images specific to hypervisors.
3. Image distribution mechanisms
- Tony wanted to follow with regard to using LFC for image distribution. He spoke with Jean Phillipe, and indeed there is a tag that could be used to indicate if an image was valid or invalid. There is no tag which could be used to indicate expiry, however expiry could be inserted as a pretext attribute.
- Martin commented that there is some complexities associated with running LFC, and it may be an imposition to smaller sites in terms of resources.
- Ewan commented that he thought that there would be only one central catalog and not one per site.
- Ewan commented that using of the LFC world offer the advantage of the WMS already understanding it. You could match a grid job against the availability of the image.
- Owen asked if we are using VO push mechanisms are we ruling out a site blocking an image?
- Tony commented that he would like to avoid sites blocking images for random reasons. For example presumably if one site wishes to block an image all sites would wish to do the same.
- Keith mention that he must reserve the right to block an image in the same way he reserves the right to block a DN.
- Keith commented that there me the case where an image is suitable to one site and not another. He cited CernVM as an example of something that wouldn't be allowed to run at BNL (because of dynamic download of configuration). At BNL, the worker nodes do not have outbound Internet access.
- Martin, Owen and Ewan will continue the discussion before the next meeting.
4. Agenda for Next week
- David will talk about security requirements for image generation.
- If time permits continue into image distribution.
5. Any other business
- Ian Gable mentioned a talk by Yushu Yao at the OSG AHM about an Atlas working group on virtualization. Tony will follow up with Massimo Lamanna. Sonder will consider getting involved with the group directly.
There are minutes attached to this event.
Show them.