Re: asynchronous multi-master replication
- From: mkb <mkb125@xxxxxxxxx>
- To: stalinsk@xxxxxxxxx, oracle-l@xxxxxxxxxxxxx
- Date: Tue, 21 Dec 2004 09:44:41 -0800 (PST)
Hi Stalin,
If your'e on 10g, have you thought of implementing
streams instead of multi-master replication?
I've implemented replication in 9i using MMR on a
couple of Solaris servers. Although it's fairly
stable, when things go wrong, they really go wrong.
One example I can give is when we accidently deleted
about 10 million rows from a table that was part of
our replication group. It hosed up the entire
replication from one site to the other. Reason being
that the deftran queue just could not handle that many
transactions in a single DML. The transaction just
sat there trying to propagate to the other site.
We tried clearing out the queues, shutting the
database down etc, but we were unable to resolve the
issue. It was finally resolved after we issued a TAR
to Oracle.
Now I haven't used streams yet, but reviewing the
docs, it appears to be implemented in a much more
scalable manner. In a nutshell, the redo logs are
scanned for DML/DDL and I believe the change vectors
or records are transferred to the replicated site and
applied there. It's not trigger based like MMR.
I'd suggest you look into Streams rather than Multi
Master Replication. I believe you'll have fewer
issues.
BTW, when it was working, we had no problems with
performance. The two sites always stayed in sync.
This was on an OTLP accounting type database.
Oh and also, DON'T use OEM to try and manage the
thing. It'll just hang as well. Their are plenty
of scripts and articles on Metalink to help you get by
w/out OEM.
hth
--
mohammed
--- Stalin <stalinsk@xxxxxxxxx> wrote:
> Hi All,
>
> We have a requirement to build an environment with 2
> databases running
> on two different locations with same set of
> applications running on each site.
> Apparently both database should be in sync all the
> time. After reading
> the doco, this requirement clearly begs for
> Asynchronous/Synchronous
> multi-master replication. Don't ask me why this
> needs to be done from
> two different locations, that's part of SLA's.
> sometimes, SLA's make
> things more complex!
>
> Can anybody share your experience in terms of
> performance/scalibility
> in a two way multi-master replication.
>
> Configuration we are looking at for now is --
>
> 4 proc, 4g mem, 15 (15k rpms) * 36 G on a raid 10,
> oracle enterprise
> edition (10g) -- per site.
>
> Thanks,
> Stalin
> --
> //www.freelists.org/webpage/oracle-l
>
__________________________________
Do you Yahoo!?
Yahoo! Mail - 250MB free storage. Do more. Manage less.
http://info.mail.yahoo.com/mail_250
--
//www.freelists.org/webpage/oracle-l
Other related posts:
