On Sun, 2005-05-29 at 10:42 -0500, Les Mikesell wrote: > On Sun, 2005-05-29 at 07:31, Alexander Dalloz wrote: > > > > > > > After experimenting with LVM, I felt that it's still got a lot of > > > > rough edges. for one, the filesystem size has to be adjusted in addition > > > > to the LVM, plus sveral other things, if you are not 100% sure of what > > > > you are doing it's easy to trash your system. It's gtting better but > > > > still needs work. > > > > > Indeed, I have the same experience too. I have used software raid > > > (Raid0) instead at the end. > > > > > SK > > > > And you want to say a RAID0 is what - a safe alternate? You must be > > kidding :) > > RAID0 certainly isn't safer, but RAID1 might be. I have most of my > critical data on software mirrored partitions and have made an effort > to arrange things so the contents fit on individual drives. This means > that not only can any single drive break and be replaced transparently > (which you could still do with LVM on raid) but I can take any single > drive, put it in a different machine, and run it or recover the data, > regardless of what happened to the rest of the original machine and > the other drives. With multi-drive LVM partitions this would require > salvaging all of the disks, making sure the recovery machine had a > somewhat matching software rev, and probably getting the drives to > detect in the same order. Does anyone have experience with this > scenario? Suppose the motherboard dies in a box with hot-swap carriers > and you've got another similar box with some empty slots. How confident > are you that you could make the LVM come up alongside one that was > already running in the other box? The drives don't need to detect in the same order; LVM and for that matter RAID using mdadm use UUIDs on the filesystems to find the partitions. I recently moved a fibrechannel raid store from a Fedora Core 3 box to a RHEL4 box. It would have been quite a smooth changeover had it not been for the fact that the raid provided two virtual drives, on different LUNs at the same SCSI ID. FC3 had no problem with this, but RHEL4 currently has to be prodded to scan multiple LUNs. Hence at first it didn't see half of the drive, and some of the logical volumes went "missing". After a mild heart attack scare, I managed to figure out what had gone wrong, fix it and get the other half of my data back. But that's a Friday evening I won't forget in a hurry :-) Paul. -- Paul Howarth <paul@xxxxxxxxxxxx>
