On Sat, Sep 09, 2017 at 07:52:19PM +0100, Anthony Youngman wrote:
> On 09/09/17 19:37, Marc MERLIN wrote:
> > On Sat, Sep 09, 2017 at 07:26:51PM +0100, Anthony Youngman wrote:
> > > But with raid 5 you can only solve for a missing/corrupt block IF you can
> > > tell the system which one is messed up. Btrfs is telling you that the data
> > > is fine, so you know it's the parity, and raid 5 can fix that for you.
> >
> > Thanks for confirming my understanding. I just wasn't 100% sure if md
> > now had extra checksums that would not allow you to reconstruct corrupt
> > data but would tell you which block was corrupted (i.e. not like parity
> > where you can actually rebuild, but enough to say "yes, this one doesn't
> > add up").
> >
> No, it's basic maths. Remember solving simultaneous equations at school? For
> every unknown you need to solve you need one extra piece of information - if
> you have three unknowns a, b, and c, then you need four non-equivalent
> equations W, X, Y and Z in order to work out what a, b and c actually are.
> Raid 5 only has one extra piece of info, so it can only solve for one
> unknown.
>
> md does have "extra checksums" as you phrase it, but that's just another
> name for parity :-) raid-5 is one checksum, raid-6 is two, and we may get
> three checksums, which I have been told might be called raid-7. Just needs
> someone to decide to sit down and write it.
I know what you're trying to say, but not quite what I meant.
You can have hash checksums over a certain amount of data, per drive.
Those checksums would cover maybe 4KB blocks (or even 1MB) and take a
mere 2 or 4 bytes. If your raid parity does not match, you can then
verify the per drive checksum and figure out which drive is "wrong".
>From there, when you repair, you know which drive's data to throw away
and recompute.
However, as I wrote this, I realized that if you lose power while
writing, you can have checksums that are consistent while the parity
data still doesn't add up, and at that point you're no better off.
Also, while those checksums don't take a lot of space, they still have
to be written in some pool somewhere, requiring reserved space.
Either way, not going to work with the current metadata format, and it
might not help as much as I was originally hoping it could.
Marc
--
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
.... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/ | PGP 1024R/763BE901
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
