Re: Q: System mirroring via archived redo logs

From: Willy Klotz <willyk_at_kbigate.stgt.sub.org>
Date: 1995/05/04
Message-ID: <799572834snx_at_kbigate.stgt.sub.org>#1/1


neils_at_pindar.com writes in article <3o5dij$9iq_at_akira.pindar.com>:
>
> Our current system (7.0.16 on Solaris 2.3) uses triggers to mirror the database
> to a backup machine. This works, but is a bit slow.
> I've been told it's possible to mirror the system by using the archived redo logs
> from the primary computer to on the backup one. Does anyone have any info (setup,
> reliability, performance, etc.) on this? Oracle are supposed to have a white
> paper on this, but I've been unable (so far) to get it from them.
>

If you get the white paper, I would like to get some info on it...

Currently we do it the following way:

Machine A runs the production database
Machine B has the backup database

At night, we copy the database from machine A to machine B; we use dd for this.

Machine A has its offline redo logs in a separate directory; this directory is periodically copied (every 10 minutes) to machine B. we use rdist for the copy.

The database has 3 online redo groups with every group having 2 members; member 1 is on machine A, member B is (via NFS) on machine B. thus, if machine A fails, we have on B

- a copy of the database from the previous night
- all offline redo-logs
- the current online redo logs.

This way, we can "startup mount" the database on B, do a "recover database", and have a current and working copy.

Also, if an online redo-log is reused, you must be sure that this log is copied to the offline redo log directory AND transferred to machine B. Otherwise, you will eventually miss a redo log (for a short time). Because of this, we use a redolog size of 20 MB, so in our setup the system must write 60 MB redo logs before we will run in such a situation (and at our site, 60MB redo logs in 10 minutes is never the case).

There is also a drawback: because the online redo logs are written to machine B per NFS, there may be a problem with your update / insert performance. But this depends on the type and load of your machine.

Willy Klotz


Willys Mail     FidoNet              2:2474/117  2:2474/118    
      willyk_at_kbigate.stgt.sub.org      VFC        ISDN/X.75
   -> No Request from 06.00 to 08.00 <-
Received on Thu May 04 1995 - 00:00:00 CEST

Original text of this message