Answers inline. On 02/13/09 10:47, Steffen Weiberle wrote: > On 01/22/09 08:22, Alexandre Chartre wrote: >> On 01/16/09 10:21, Tom Gendron wrote: >>> >>> Liam Merwick wrote: >>>> On 16/01/2009 15:29, Tom Gendron wrote: >>>> >>>>> Hi, >>>>> >>>>> I am playing with the warm migration feature of ldm 1.1 and notice >>>>> that >>>>> there is no provision made to move the root disk from source to target >>>>> machine. I must do this by some external means correct? >>>>> >>>>> >>>> Yes, the disk backend(s) must be visible to both machines. e.g. a >>>> SAN or >>>> have the disk image files exported via NFS. >>>> >>>> >>> I see how NFS could do this but I am confused on how to do it with my >>> SAN. I have direct attached fiber channel disks that are visible to >>> both hosts but I am not sure how to share the file system. ZFS is >>> going to label the pool and be tied to one of the hosts. I would have >>> to explicitly export it and then import it to the target machine. Is >>> there a filesystem solution to this with native Solaris tools? I >>> think QFS would solve this but that is not native. Anything else? >> >> With a SAN, you directly export a LUN as a virtual disk, and the guest >> domain will use that disk. Then you can use ZFS or whatever filesystem >> from the guest domain. > > Hi Alex, just a clarification regarding the file system on the LUN. Via > a statement, although really a question :) > > Any single writer file system can be used as the file system on the LUN. > The source domain will initially be reading and writing to the LUN, > while during the migration the target domain will be reading only (if at > all), thus there is no issue of file system cache conflicts.
During the migration the source domain is suspended so it is not doing any I/O, and the target domain is not running yet so it is also not doing any I/O. > Once the source domain has been suspended completely, the target domain > takes full control of the LUN. So whether the file system in the LUN is > UFS or ZFS, there is no concern about lack of file system cache mismatch. Yes, you don't care about the type of filesystem because any I/O operation will be suspend on the source and then resumed on the target. >>> How are we planning to solve this when we complete the migration >>> capability as in live migration? >> >> We plan to support more types of storage, like ZFS in some future version >> of migration. Basically a zpool will be initially imported on the source > > This is the case where the boot or data devices are vdevs or files on > ZFS, which is separate from whether ZFS is the file system as understood > by the OS running in the guest domain. (I am trying to differentiate how > Tom's question could be interpreted--how I am trying to distinguish my > interpretations of it! ) Yes this would be the case where the vdisk backend is a ZFS volume or file from a zpool that can be imported/exported between the source and the target. Then you can put whatever filesystem you want on the vdisk itself. alex. >> system, then when the target domain is suspended the zpool will be >> exported >> from the target system and imported on the source system. Finally the >> domain >> will be resumed on the target system. So the zpool will be automatically >> migrated during the migration by using a zpool export/import. >> >> alex. > _______________________________________________ > ldoms-discuss mailing list > ldoms-discuss at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/ldoms-discuss
