On Thu, Feb 01, 2007 at 01:06:54PM -0500, Michael S. Peek wrote: > So the consensus seems to be that LVM is the way to go. > > So what's the cutoff between building arrays of varying size versus > grouping them under LVM? > > I.e. Right now I've got two large arrays. Should I maybe break that > down unto just a bunch of disks and then use LVM to group them together > (not use hardware RAID at all), or should I break the disks into each > bundles of three and make as many small raid5 arrays as I can and then > group them under LVM? well the thing with LVM is if you have a disk failure one layer below the lvm then you've likely lost a bunch of data. Only if you're lucky enough to not have any extents on the failing disk will you not lose data. Fat chance that. By using RAID you are moving that disk failure down another layer from the lvm. A single disk failure when you have jbod under lvm means your toast. A single disk failure when you have RAID under lvm means you have and intact lvm running on a degraded array -- a much preferable situation ;-) I think the great utility of lvm is not in the combining of multiple volumes into a big volume group but instead in the flexibility of being able to shuffle around your file systems as needed. So I wouldn't give up the reliability bonus of RAID underneath lvm. That way if you have a disk in your array fail, you still have a valid lvm setup over the top of it. .02 A
Attachment:
signature.asc
Description: Digital signature