HEPiX Virtulisation WG meeting minutes 2010-06-07

Present (in no particular order):
Tony Cass, CERN
Romain Wartel, CERN
Ewan MacMahon, Oxford
Martin Bly, RAL
Thomas Finnern, DESY
Owen Synge, DESY
Ian Gable, University of Victoria
Michel Jouvin, GRIF
Abdeslem Djaoui, RAL
Sebastien Goasguen, CERN
Jason Shih
Sonder Klous, NIKEF

1. Comment on the previous meeting minutes

• Dave has not had a chance to distribute the policy document.

2.  VM Image generation

• Andrea has succeeded in creating an image able to boot as KVM, Xen HVM, and Xen para-virt, by passing arguments to pygrub (bootargs="--entry=n"  where n is the desired grub entry.)
• Andrea will provide a recipe for creating images of this type.
• Andrea was not able to make any performance tests with virt I/O, however he suspects that virt I/O will be a possibility for the final image.
• Michel Jouvin commented that kernel models could be part of contextualization.

2. Contextualization

Sebastian went over slides that he prepared on contextualization:

• He defined two types of contextualization: image contextualization(modification of the fs prior to boot), and instance contextualization.
• He recommended that we use instance contextualization.
• He recommended a single file in raw format as the image type.
• According to the OVF standard, Instance contextualization is to be done via an attached ISO image, and he recommended this method.

Discussion on contextualization   

• Owen commented that mac address should not be stored in the image, and Sebastian agreed that this was consistent with the proposal.
• Abdeslem asked how images information can be exchanged with ISO images i.e. for contextualization of groups of images into self contained clusters.
• Sebastian commented that you can't achieve the self contained mini cluster with the ISO image method, but you can achieve a cluster by having the images contextualized to connect to something like and LSF batch queue.
• Tony commented that we can limit the scope of the group's efforts to contextualization of single images, which fits well with having VM connect to a site fabric.
• Tony proposed that the group focus on using the attached ISO image, and the group agreed.
• Ewan proposed that we allow the ISO Joliet extensions, the group agreed.
• Sebastian commented that at CERN the contextualization occurs before networking starts.
• Ewan commented that he thinks that having all images configured for DHCP would be adequate.
• Ian Gable commented that he felt it was useful to have access to contextualize networking.
• Tony commented that that having a contextualization script run as the very last step in the boot up process would allow the contextualizer the last word in the configuration..
• Tony commented that it would seem that there should be two contextualization scripts one run immediately before the start of networking and one run as the very last step in the boot up process.
• Sebastian will come up with a proposal for the name of the contextualization scripts within the ISO images.

Image format sub discussion:

• Ian mentioned that he had a strong preference for easily mountable disk images for the purposes of debugging and investigation. Ian asked if there was anyone else who shared this preference.
• Owen also shares this preferenc. He sighted the example of having difficulty mounting a hard disk image containing LVM partitions.
• Both Tony and Ewen thought that it could be unnecessary to impose restrictions on the format of the disk image.
• Tony asked that this discussion be suspended until a later meeting.

3. Image Distribution

• Last thursday the image distribution group met, however Martin was unable to attend.
• They wish to have some experts review their database design.
• Ewan had a chance to look further at the 'straw man' implementation.
• Ewan has  an X.509 certificate for the service but has not yet picked it up.

4. Next meeting

•  Friday the 18th 14:00 UTC