> I've been using this scenario for database backups since we had the > discussion about shared VM disks on this thread about a year ago. > David made the point then that if you had a hipervisor that allowed it, not > making a disk dependent on one of the machines being up at all times (as > with NFS), had some advantages.. I also used Tore's suggestion to stick to > ext2 so nobody would try to replay a journal. .I use them for database > backups that I take each night, and for SLES10 (both SP1 and SP3) it works > fine. IF the RW linux is down (usually because it didn't come up again), > caching won't be an issue.
The key issue is that the filesystem inside the Linux guests must coordinate access between the machines. Sharing a disk with ANY writers involved using a non-cluster aware filesystem WILL corrupt your data at some point. You also should specify NOMDC in the CP directory entry for shared disks where there may be any writing done by any of the sharers. R/O for all is fine -- that's why sharing things like /usr that rarely change works well. You *really* don't want to do this for user data that may change. Side note: Look at XFS instead of ext2 if you're feeling lucky and decide to do this. It has some tooling designed for use with ROM-based filesystems that helps postpone when you'll have to clean up the mess. > I haven't tried sharing a disk between 2 SLES 11 machines yet, and > that's where I'm going with upgrades. If it doesn't work, it would be > nice to know. Why would the fact that one of the machines has the disk > mounted RW keep the other(s) from seeing it at all? The point being made is that while it can be configured, it WILL screw you at some point. The two virtual machines don't know that each other are present; they can't see any caching that happens because it doesn't know the other machine is there. Any time you modify the shared disk, the other machine won't pick up the changes unless you have some software coordinating that access. Your second machine won't see the update and will merrily serve up corrupted data. This leads to Having To Explain What Happened -- generally a bad situation. Linux has a number of good cluster file systems: GFS2, gLustre, Ceph, PVFS2, etc. You could even use OCFS2 if you can deal with Oracle not distributing the current version on anything but their distribution. But, use SOMETHING that knows about multiple machines if you're going to share a single disk. ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/
