Re: 24x7 with full backup

GA,

You need to define your recovery scenario 100%. Does 30 minutes mean all users are back online, or the first user has the ability to be back online. Do you want the database simply restarted on the same machine (Crash Recovery) or do you want to be able to recover from disk failures in the 30 minute period (Media Recovery), or do you want to be able to recover from a complete machine or site failure (Disaster Recovery).

If you are wanting Crash Recovery, then Oracle8 should be OK without any modifications, except for Log Check pointing needing to be reduced to about 10 minutes. (log_checkpoint_timeout = 600 (time defined in seconds between checkpoints)).

If you are wanting Media or Disaster recovery, it sounds like you may be needing Oracle's Standby Server.

With Standby Server, the Oracle Redo Logs are copied from the master database when they are archived, and the standby database is in constant database recovery. The Standby database can be on the same machine as the primary database, but this does not provide resilience to a machine failure, so to have full Disaster Recovery, you need the standby server on a second node. This does have it's own problems, apart from cost. If you are capable of loosing the current redo log contents, then you can complete recovery after applying the last archived redo log files. If however, you need to have the last redo log file (if possible) then you need to have dual mountable disks for the Online Redo Log files, so when the primary machine crashes the online redo log file can be accessed by the standby database.

These are the three methods of normal recovery supported by Oracle. However, if you are wanting to be resilient from even block corruption (of the redo log files) then you need to go to a replication approach, and again you need to think what level of resilience / loss of data you are prepared to accept.

The Oracle Backup and Recovery manual will talk about Crash Recovery and Media / Disaster recovery using Oracle's Standby Server, and it might provide some sort of suggestions on Replication Server.

Hope this helps, but as I am sure you understand, recovery is specified in Dollars. To go from 99% availability to 99.9% availability it costs about 1 unit. To add an additional 9 (i.e. 99.99) multiply this by 10, and you are still only at about 1 days down time per year. To get to 99.99999% you are looking at millions of dollars to achieve the 10 minutes per year down time offered.

Andrew

ga wrote:

> My company just made a decision to go 24x7 with oracle 8.0.5.
>
> I am stuck with finding out if there is an easy way (i know that within
> oracle nothing is every easy) to get an oracle server up with 24 hours -
> 7 days a week access. And if the server dies, be back up - with minimum
> data loss within 1/2 hour.
>
> I have heard about oracle 8 with enterprise server and replication.
>
> Does that sound like the right path?
>
> I have no experiance with replication and do not know what its
> performance is or how complicated it is to:
> set up
> manage
> cut over to backup if the primary is down
> go back to primary when it is up to date
>
> I know this is a long question - sorry.
>
> thanks ahead for any info
> GA
Received on Mon Apr 19 1999 - 07:50:19 CDT