Re: Cunning Plan

If you are truly mirroring between the two SANs, this should work most of the time. It is no different than a disk array that drops a single mirrored disk from an array. However, if there are any synchronization issues, you could have a partially valid instance, meaning your control files might be correct with invalid data files or vice versa. The chances of an unrecoverable media error are non-zero. Your hardware might be able to reduce the likelihood of this type of error, but it could still occur.

True replication or a true standby database from logs should never result in an invalid instance that you cannot bring online if the primary dies. Your scheme seems to contain some operating system and communication errors that could sabotage your primary instance.

You may have to spend the "quid" to test this in order to validate the model. However, the tests will have to be pretty brutal, meaning forcing physical disconnects in the middle of large transactions, and you still might not be able to reproduce a production case failure.

Good Luck!

--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Chris Weiss
mailto:chris_at_hpdbe.com
www.hpdbe.com
High Performance Database Engineering
Available for long and short term contracts


"Ralph" <rlro99_at_hotmail.com> wrote in message
news:e2c49cae.0205030448.34b12856_at_posting.google.com...


> Hi all,

>

> We have a cunning plan for our architecture and I am interested to

> know if anybody can see any obvious gotchas before we go any further

> and I end up looking a bit of a Baldrick?.

>

> Our prod system (8.1.7.3) currently runs on a cluster of two E6500s

> using Veritas Cluster Server using a shared hitachi HDS9900 SAN. When

> one of these boxes has a "Microcode Error" and falls over, the other

> box picks up and we are running quite nicely again in about 5 mins. In

> addition to this we run another E6500 in another datacentre about 50

> miles away and ship logs to a standby database for diaster recovery

> purposes. All this works well, but everytime we are required to

> increase capacity we need 3x CPU and 2x Oracle Licence.

>

> We want to change this so that we drop the second machine from the

> cluster and cluster to the DR machine instead. A "Dark Fibre" link has

> reacently been installed between the two data centres and we aim to

> replicate the data between the two SAN's using this link and do away

> with the standby database. In the event of failover presumably we can

> just startup the instance on the DR box and reverse the replication

> between the two SANS.

>

> This will save us hardware, software and eliminate the potential data

> loss in activating the standbay database?

>

> What do you think? This will involve us spending quite a few quid just

> to set up for testing purposes, I can't see a problem...

>

> Thanks for your time.

>

> Ralph.