Re: Multiplexing Datafiles?

There are three issues involved. Backup, Redundancy for purposes of load sharing, and Redundancy for purposes of failover and disaster protection. Howard is right about backup.

Oracle offers three choices for redundancy: a standby database, parallel server, or advanced replication.

    A standby database does not lend itself to either load balancing or true failover. It does lend itself to disaster protection since the primary and standby can be physically separate. The standby is configured like the primary. As Oracle archives its online redo logs they are sent to the standby which applies them. Thus the standby is always behind the primary and almost never has all the changes which have been made on the primary. Anything in the primary's online redo logs that have not yet been archived, hasn't even been sent to the standby, let alone applied.

    When the primary fails, as much information as possible is applied to the standby, which is then brought online. This takes time and some data may be lost. In addition the standby server, which is as expensive as the primary server, does no work - it just waits for the primary to fail. And until you repair or replace the primary, the standby has no standby.

    Of course all the information that flows from primary to standby moves across the network. If the machines are close enough to use some non-network means of communication, then they may be vulnerable to some other form of failure - lightning, intentional damage, etc.

    Parallel server creates multiple instances that run against a single set of datafiles. This implies a single set of disks, which removes some of the redundancy that's useful for disaster protection. Also, the instances run on what Oracle calls "a cluster or massively parallel hardware platform". As with a standby database, a parallel system is vulnerable to some type of site failure - lightning, intentional damage, etc.

    While the processors share the load of doing whatever can be done in memory, all instances are reading and writing to a common set of disks.

    Failover is achieved from the fact that instances can be added and removed from the parallel server environment. If an instance fails, the other instances keep running and working. If loads on the system increase, new instances can be added.

    The third method is advanced replication. In this method two or more identical servers are configured in the same way with each containing identical databases and each having its own set of disk drives. All servers can be used to do meaningful work, so load balancing is a real possibility. Changes made to the database on any server are replicated to the others - either immediately or with some programmed delay. If a server is lost, the others can continue to process requests. And the servers can by physically separate, although that implies a lot of update traffic over the network.

    This method sounds promising, but I could list several pages of potential pitfalls in its actual application. So you'll have to go to the Oracle advanced replication manual for the gory details.

Van

"Howard J. Rogers" <howardjr_at_www.com> wrote in message news:3995df0c$1_at_news.iprimus.com.au...
>
> "Rob Jolliffe" <rjolliffe_at_my-deja.com> wrote in message
> news:8n4295$mqp$1_at_nnrp1.deja.com...
> > Hi,
> >
> > I'm just beginning to learn Oracle, and I was reviewing some Email
> > regarding Raid5 vs Raid0 performance. It seems Raid 0 is significantly
> > better for performance - but you lose the redundancy of Raid 5 (or even
> > Raid 1).
> >
> > I was trying to also determine if in a data tablespace it is possible to
> > multiplex (mirror) datafiles the same way you can with a Redo log group.
> > I was totally unable to determine if you are or are not.
> >
>
> Absolutely not. Never has been, and there never will be -because it is
> entirely pointless. The reason for duplexing your data files is,
> presumably, to avoid complete disaster in the event that one of them dies,
> gets deleted, corrupted etc. That is precisely what the Redo Logs are
 there
> for. Provided you've kept all your redo since the time of the last
 backup,
> you can reconstruct any given data file up to the precise time of failure.
>
> So if you're mirroring your Redo, you have provided all the protection
 your
> data files ever need.
>
> (Conceivably, you can use external, hardware mirroring of data files to
> enable recovery to take place *faster* -you just break the mirror and use
> the good copy. No need to wait for all that Redo to be re-played. But
> that's a mean-time-to-recovery issue. It's certainly not necessary).
>
> > Can anyone tell me? I know you can stripe tablespaces - but can you
> > mirror them? Perhaps this requires creating two tablespaces and somehow
> > matching them?
> >
>
> Forget it. There's no such concept in Oracle, because the Redo does all
 the
> protection work we need.
>
> Regards
> HJR
>
>
>
> > Thanks very much
> >
> > -Rob
> >
> >
> > =============
> > From the desk of Rob Jolliffe
> >
> >
> > Sent via Deja.com http://www.deja.com/
> > Before you buy.
>
>
Received on Sun Aug 13 2000 - 00:00:00 CDT