On Wed, 15 May 2002, John Orthoefer wrote:
> Yep this is a classic mainframe backup technique.  You make a triple
> mirror (Master + 2 mirrors) then when you want to back up your database you
<snip>
> That way you still have your mirror with current pending transactions
> even if you are in the middle of doing a backup.  And the downtime for
> the backup can be just a few seconds.   This was used in the day when
> RAID systems wheren't that common.  Today I would use either 2 raid
> systems, or even a raid system, and some JBOD for the backup mirror.
<snip>
> Netapp has the same functionality, it's call snap mirror I think.
>  Again, you are talking about spending a bit of cash with netapp.  They
> should give you engineering support to make it work.

okay, so i'm a little behind the discussion...

netapp's snap mirror accomplishes what the OP wants, but in a much more
disk-efficient manner than other suggestions (as i understand them):

 netapp is usually deployed in a raid 3 configuration.  i'm not even
 sure if there's a raid 1 option, so cut your disk requirements in
 half right there...

 the snap mirror is *not* a copy of your data, though it appears to be.
 it's a snapshot of the filesystem metadata, (inode and superblock type
 things).  the data that's mounted in the snapshot directory is the same
 data blocks you're working with in the live filesystem...  no extra
 copies (save raid 3 parity).  as you change data in the live
 filesystem, the changed blocks are written to free areas of the
 filesystem, and to the live fs metadata.  the snapshot (using old
 metadata) is unchanged, and the data blocks used by the snapshot aren't
 overwritten until the snapshot is deleted.  there's no mirror to resync
 because there's no second copy of the data.  => even more disk savings.

let's say your filesystem size hangs out around 1TB, but you change
around 1GB every hour...  with a 1.25TB (after raid parity, hotspares)
filer, you could take a snapshot every hour for 10 days before you run
out of space!  all would be mounted, and available, just like the OP
wants them...

having that many online copies with the other solutions would require
hundreds of TB.  ...and the netapp solution doesn't have a heinous
resync time that the others have when the mirror is restored.

of course, you can't really use a netapp filer for a database and
expect any sort of reasonable performance.  but they make one hell of
a good fileserver.  i love 'em.  can you tell?

/chris


---
Send mail for the `bblisa' mailing list to `[EMAIL PROTECTED]'.
Mail administrative requests to `[EMAIL PROTECTED]'.

Reply via email to