Re: Standby strategy

We have two machines (HP) in a failover cluster using HP's MC ServiceGuard. The disks are 'twin-tailed', that is , the disks are connected physically to both machines. The MC ServiceGuard software sees that only one of the machines at a time 'mounts' the disks. If the database machine fails, MC ServiceGuard 'dismounts' the disks from that machine and 'mounts' them to the spare and re-starts the database. Note that this does NOT happen if just the database fails.

John

Nigel Rudgyard wrote:
>
> We are currently trying to devise a suitable strategy for maintaining a hot
> standby system using two clustered DEC Alphas. Two disks within each box are
> volume shadowed thus giving each machine visibility to all the Oracle files
> within the installation. We are using Oracle v7.1.5.
>
> Up until recently we had been intending to make use of the Oracle Parallel
> Server option, however due to the associated costs, this solution is no
> longer acceptable to our client unless it represents the viable option
> within the project timescales.
>
> Since we are not really using OPS as it was originally intended - for the
> most part the standby machine is idle - it crossed my mind that a much
> cheaper solution is potentially possible. Assuming the two machines are
> known as A and B, the strategy goes something like this; a single database
> instance starts up (in exclusive mode) on machine A. At failover the Oracle
> instance upon A is shutdown (assuming that it is still up) and a new
> database instance, accessing the same database, in started up upon B.
> Simple.
>
> Now so long as we can guarantee that the two instances (or more precisely
> the two occurences of the same database instance) are never connected to the
> database simultaneously then we are OK. If this cannot be guaranteed then
> the database is in danger of being corrupted. Note that unlike two instances
> running upon the same machine I don't believe that Oracle will trap this on
> our behalf. The 2nd instance to connect will, after interrogating the
> control files, presumably see that there is an inconsistency and attempt to
> perform crash recovery.
>
> Has anybody had any experience of implementing such a strategy ? Is it
> viable ?
  Received on Mon Mar 16 1998 - 00:00:00 CST