On 3/24/10 3:02 PM, "Mark Post" <[email protected]> wrote:

> This whole setup seems somehow wrong.  I've always seen references to linking
> minidisks MW as being a sure way to guarantee data corruption/loss.  I suppose
> that's because of the minidisk caching (MDC) that CP does, but I'm not sure
> that's the only reason.  Perhaps it's a CMS only issue, but I sure wouldn't
> want to bet my business on it.

Historically, it was because CMS cached the FST (the map of allocated blocks
on the minidisk) in the storage of the virtual machine and made no effort to
sync it between virtual machines (because, well, they're separate machines,
and one shouldn't interact with another unless you ask for it). Update one,
and bang, all the rest are wrong and the next write (like updating last
access date) destroys the consistency of the FST on disk. Similar problem
here; OCFS does not inform other virtual machines which one has control of
the resource and manage the writes appropriately.

> Additionally, LVM in SLES10 is not cluster aware.  In the SLES11 High
> Availability Extension, cLVM is, but in SLES10, only EVMS can provide that.
> Are they using EVMS, or LVM?

(OT: Remind me again why LVM was a better choice than EVMS, Linus? *sigh*
Not your fault, Mark, but it's a PITA to have to teach the Linux world how
to do shared storage and make all the same mistakes again. )

> If they really want to stay with SLES10 and are using LVM, I think they need
> to move to a non-LVM SCSI over FCP setup.

If for support comprehension reasons if nothing else. CLVM still doesn't
handle the MW minidisk case correctly, and if you call in for support on
it, you're going to get a lot of very confused people on the other end of
the phone. 

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to