Re: Redo + Archive + Data Layout...

"Howard J. Rogers" <hjr_at_dizwell.com> wrote in message news:<3f950534$0$10411$afc38c87_at_news.optusnet.com.au>...
> Daniel Morgan wrote:
>
>
> >>
> > You are correct. I forgot that Hari seems to have a compulsive need to
> > re-ask questions as though they
> > were never answered. But I disagree about mirror what has been
> > multiplexed. I can't think of a possible
> > failure that would be better solved by having 4 control files than 2.
> > The issue is eliminating a single point
> > of failure. If you had two corrupt control files ... your hardware
> > mirroring would just give you mirrored
> > corruption.
> >
>
>
> Hardware mirroring protects against hardware failure; software multiplexing
> protects against software corruptions and user errors.
>
> If I rm control*, hardware mirroring will not save me, because the rm will
> be replicated on the mirror. Or if CKPT is updating the controlfile, throws
> a wobbly, and introduces corruption, that corruption will be mirrored.
>
> Only by making CKPT write to two separate and independent multiplexed
> versions of your controlfiles can you protect against that sort of error.
>
> Of course, having two multiplexed copies means (maybe) you don't need to
> bother with the hardware mirroring in the first place. And if money and
> resources were tight, I'd go for multiplexing over mirroring every time.
>
> But there's one thing hardware mirroring can do for you that Oracle
> multiplexing doesn't do: splitting mirrors for the taking of backups.
>
> It's also true that in a world of 9 possible archive destinations, 2 control
> files is not sufficient protection. So why not Oracle-multiplex 3, 4 or
> more times? Because doing so causes CKPT to write to more files (and LGWR,
> and other processes), and you've therefore increased Oracle's workload.
> Having 2 files managed by Oracle, and then mirrored by the hardware, gives
> you the larger number of files you may require for resilience and peace of
> mind, without increasing the workload for Oracle unacceptably.
>
> So, if I had the choice, I'd multiplex and then mirror. Best of both worlds.
>
> Regards
> HJR
Howard,

the other concept is, as one adds more controlfiles, one introduces more points of failure. Loss of a controlfile causes instance crash.

This is similar to adding more drives to a RAID-F volume. The larger the number of drives, the smaller the MTBF for the volume.

So it would fall under the concept of diminishing returns, whereby it can just be safely proclaimed:

"Three is the number, the number shall be three." And all agreed, and it was good.

maybe.

Pd Received on Tue Oct 21 2003 - 10:53:41 CDT