On Wed, Sep 17, 2003 at 02:48:10PM -0400, Kent Borg wrote:
> How long does it take to do a backup of the 3.5TB storage? How long
> does it take to get a file out of the backup? How long to do a
> restore? This is different from home network territory.
You're trying to make me cuss again aren't you? :-). That 3.5TB system
happens to be Windows with 450GB-500GB file systems (no I did not, and
would not, design it this way). Backups take between forever and
forever and a day. To the best of my knowledge, the admins have never
attempted a full restore of a 500GB file system and I know it would not
be a pretty sight. Individual file restores are reasonably quick,
though, and we do these regularly.
I've done extensive testing with backups on my VMS cluster using
multiple backup packages. This cluster has about 800GB of usable space
and is one of the ugliest backup environments known - there are over 15
million little files on that sucker. The cluster was put together to
simply not lose data. All controllers are redundant within data centers
and across data centers - 2 pairs of controllers in each data center
(across the street from each other). We use RAID-5 within each data
center and RAID-1 across data centers. I have seen double-disk errors
in a RAID-5 set - more than once - and not lost data. Those of who
think that RAID-5 with a hot spare is good enough may see realize
differently, but for your sake, I hope not. Remember those geriatric
jokes on this list a few weeks back? I was taken 'em...
On my VMScluster, I can currently find files in my catalog in under 10
seconds and it will identify the tape and the file mark on tape. To
complete the restore is usually an hour or two.
We just started running our backups through NetBackup, and although the
backups are faster due to increased parallelism, restores suck - over 2
hours to find a single file in the catalog. That's an issue I'm working
on and could be a show-stopper for our migration. The catalog is
already over 2GB and we just got started.
We've done terabyte database restores and they're rather quick. Our
storage subsystem is definitely high-end with EMC Symmetrix controllers,
redundant HBAs, with a separate SAN just for backup traffic.
> More questions: How much data changes from one backup to the next? Is
> there room for incremental backups, a la rsnapshot?
Oh yeah. The classic design problems are here - the vast majority of
the activity happens on the tiny minority of the disks and really not
that much data is changing that often. On the 3.5TB system, rsync would
be a great solution. On my VMScluster, not as much so since the data
does change much more regularly. I've got one 50GB disk with over 4
million tiny files on it.
I know that there's room for improvement in our current backups, and
expanding on ideas that I've already seen point to some nice
efficiencies that I can do.
Based on the data, we do mix differential and full backups on varying
schedules. Some disks that only grow get less frequent fulls, and more
frequent differentials.
> -kb, the Kent who wants to know how the big and medium guys do it.
.../Ed (who wants to work on smaller systems some days!)
--
Ed Wilts, Mounds View, MN, USA
mailto:[EMAIL PROTECTED]
Member #1, Red Hat Community Ambassador Program
--
redhat-list mailing list
unsubscribe mailto:[EMAIL PROTECTED]
https://www.redhat.com/mailman/listinfo/redhat-list
