Hi, all –
I’m faced with a pretty unique situation that I can’t
quite think of a way in which Bacula can effectively back up exactly what I’m
trying to back up.
The ultimate goal is to be able to run incremental,
differential, and full backups.
Playing around with some performance numbers and we’ve
found at least a 10x speed improvement when using an iSCSI target from our
VMWare host as a raw device for the VM, over a file-based device for the VM
which would reside on that same VM host. Although that’s great, I’m
still not comfortable doing that given the fact that we can’t reliably
back up the data.
I’m looking for an effective way to back up any and
all of these VMs, and the problem resides with the method by which we are
actually using the target. The original idea was to use LVM to take
snapshots and thereby manage incremental and differential backups.
However, VMWare – using that iSCSI target as a raw device – would leave
that device as type LVM, but rather, whatever filesystem that the host
uses. Even if that wasn’t an issue, it would still defeat the
purpose of using iSCSI devices, since VMWare would then have to be forced to
create a file-based filesystem for the guest. Catch-22, you know? To
make things even better, we’re going to have quite a few Windows VMs,
too.
I’d rather also not use bacula-fd from within the
vmware guest, as I don’t want to double-up on the iSCSI path usage –
however, this may be our only choice. If I can simply backup from the raw
device and also find a way to do incremental and differential backups off of
that, that would be quite ideal.
It looks like right now the only way to actually get a
snapshot of that device (rather, vmware disk) is to use some proprietary
software by VMWare. Although I’m not completely opposed to that, I’d
like to stick with Bacula, iSCSI, and some good old fashioned hackery.
Thanks
-dant