SUMMARY: advice needed on EMC box as one large filesystem

From: Harald.Knipp@bfa.de
Date: Wed Sep 04 2002 - 11:56:23 EDT


Hi all,

first, thanks to

Bryan Lavelle from HP
Bjorn Ramqvist and
alan

In short I asked the list for comments on a SAN configuration
with two EMC boxes and a AS1200 with Tru64 5.1.
The EMC boxes export 72 Disks to the AS 1200 and the backup-people
asked me to put all volumes into one filedomain (see my original posting at
the end).

Its obvious (and thats what all who replied pointed out) that this
configuration
could not be called 'reliable'. The point is: loosing one disk in a
multi-volume
file domain means loosing the whole domain!

Statements were:

>From Bryan:
[..] the AdvFS guide recommends that no more than 10 volumes
be placed in a single domain, for the very reason you state, that
the possiblity of losing a domain increases with the addition of
too many volumes.
[..] You can use LSM to mirror across the boxes.

Using LSM was my first idea, but first, we need about 1.5 to 2 TB space,
and if
we double-mirror we have only 1.25 TB left and second, this is just
an interim solution for a few months, so they didn't want to spend money on
a LSM license.

Bjorn brings it to the point:
"[..] but I just think that's insane!"

and from alan:
[..] When data availability is the primary concern, it is
quite common to use multiple levels of mirroring. [..]

Thats what I read in many other sources. On a first
sight two or even more levels of mirroring look like excessive waste of
disk space, but one should have in mind that in a worst case
just "recreating the domain and restoring data from backup" doesn't
help because it _is_ the backup were talking about.

In the end I found out that the information from the backup folks
on the RAID level of the volumes in the EMC was just wrong.
The EMC tech configured the volumes as something they call a RAID S.
It's similiar to a RAID 5, with the difference that you can loose more than
one disk and the volume is still availible.
As I said, it's just an interim solution so I made a compromise with the
backup
people. We put the volumes in two file domains, one for each EMC,
make alternate backups in both domains and trust on the legendary
reliability of the EMCs.

Greetings

Harald

My original posting
----------------------------------------------------------------------------------------

Hi all,

perhaps some of you remember my question two weeks ago
concerning the possibility to restore a multi-volume
advfs domain with a crashed disk. The problem was in short that we tried to
dd the reactivated
crashed disk to the new disk. The attempt to restore the domain then
crashed the whole server (AS 2000 with 4.0e).
(See my summary:
http://www.xray.mpe.mpg.de/mailing-lists/tru64-unix-managers/2002-08/msg00248.html
)

Not only since then I´m a little paranoid with multi volume AdvFS domains.
If you have
n disks in a domain, for example, mounted on a single mount point,
the possibiliy to loose the data in a whole file-system grows at least by
n.

And now the real nightmare comes to me:

Since last week there are two EMC (Symmetrix 8450) Boxes connected to one
of
our AS 1200, running Tru64 V. 5.1a.
The configuration of the EMC is: all disks mirrored (RAID 1), 4 disks make
up a logical volume
(stripe in EMC speak, I think) of 32 GB. All logical volumes are connected
to the AS 1200 via a
KGPSA FC card (two connections to each box).
Now infinite horror starts: we have 78 (!!) disks from the EMC now (about
2.5 TB)
and the backup people want me to put _all_ disks
in one single advfs file domain so that they can see the whole storage
under one mount point. The reason (they told me) is that their backup
software cannot handle
multiple mountpoints as one backup area. (BTW: the backup software is
sesam2000).

Questions:

- Am I to paranoid (should I take Prosac)? If not:
- Any ideas or hints to make this configuration safer (e.g. by using LSM)?
- Is it easier to restore a crashed multivolume domain with the new AdvFS
version?
- Any experiences with the above mentioned EMC configuration?
  How reliable are these boxes in such a configuration?

TIA,

Harald



This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:48:51 EDT