Steve Litt <slitt@???> wrote:

> You just reminded me why I don't use RAID.

I find that statement interesting. What do you use instead - rely on backups ?


md on Linux is actually quite easy to use. OK, it takes some getting used to, but after a few goes it’s not too hard. Plus it’s very flexible - the hardware raid systems I’ve used in the past certainly didn’t allow the sort of operations I and others have described.

I’ll add that my viewpoint is coloured by having had my backside saved by raid so many times. Yes, if a disk fails you can replace it, restore your backups, and carry on. But in a business world that takes on a whole new meaning :
Firstly, there WILL be lost data, typically everything all your staff did today since last night’s backup.
Secondly, you’ll have people sat around twiddling their thumbs while to get everything back up again - just the restore typically takes “hours”.
In business, there really is no case for not using raid on anything but things like mass desktops/laptops that can be rebuilt easily and which don’t hold any data.

And at home, I have raid on my Linux boxes. I just don’t want the hassle, lost time, and lost data from having to restore (e.g.) my mail server (or rather, the host it’s a VM on) from a backup if a disk fails. And I have had disk failures, but I’ve been able to replace them and do a hot rebuild without any downtime or lost data. That’s been well worth the extra cost of hardware, and the time taken to set it up.

That’s not to say I don’t have backups as well, raid only protects you agains a failed disk - not again other mishaps such as “oops, that wasn’t the directory I intended to delete”.

Of course, the operations I and others have suggested have come about from a desire to repartition disks into different arrays. That’s not a common requirement - and again, the fact that with md it’s possible to do so while keeping your data on the internal disks is almost magical compared the other raids I’ve previously used where the ONLY option would be to nuke the array and your data before building new ones.

And a way to avoid the requirement the OP asked for is, as I do, to add LVM on top of the array - though I think LVM can handle the raid as well now ? So two levels of abstraction between partitions on disk and filesystems - I would say in the Linux world we are blessed with this richness of capabilities.

And I confess I keep thinking I should have a proper look at ZFS - but I just never find the time, and I already know md + lvm “well enough” for my now fairly modest needs.


Obviously that’s my viewpoint, YMMV.


Simon