SAN, Tru64UNIX, Continuous Access (CA) and Oracle integration

[Leonid Khrisanfov]

I have two cluster systems:
1. 2(two)AlphaServer DS25 + 1(one) EVA5000
2. 2(two)AlphaServer DS20E + 1 (one) EVA5000
All clusters are running under Tru64UNIX v.5.1B
with TruCluster Software. It is not 4-nodes cluster, but 2 separate 2-node clusters. Production application - Oracle 9i RAC. I have Continuous Access feature licence.

I'm going to implement following solution:
The first cluster is a primary Oracle server and the second cluster is a Oracle stanby server for read-only transactions. Is it possible to use Oracle archivelog mechanism + CA feature via SAN to keep secondary DataBase consistent?

The fact is that mirror disk (DR group) at second EVA storage is presented to remote cluster hosts but they cannot to mount it until failover occurs. But I need readonly access of second cluster to mirror disk.

Any advises?

[Uwe Zessin]

Leonid,
I don't know if CA allows a way that the target can be seen read-only (from what I have heard so far I think not), but even if it were possible I strongly suggest that you do not try this.

The system that has the disk mounted read-only will not be aware of any changes to the filesystem - there is no synchronization.

I suggest you use log-shipping to keep your standby database current. You can still use CA to have a recent copy of the redo logs at the target which can be applied manually after a failover.

[Peter Mattei]

Do not use the target DR group while replication!

You said that you do use a standbyserver where logs are shipped to. Why do you not use that data?

If you need to access production data volumes you could create a snapshot or snapclone (point in time copy) of the volumes using Business Copy EVA. This has to be don on the primary side!

The CA target side is only to be used after a failover or after a "suspend" command has been issued.
If you suspend your DR group the primary EVA starts logging all writes. The DR group will be resynchronized after a "resume" command.
While suspended you can use the target voulmes for things like backup or data warehousing.

I suggest you have a look at the CA Operations Guide
http://h20000.www2.hp.com/bizsupport/CoreRedirect.jsp?targetPage=http%3A%2F%2Fh200002.www2.hp.com%2Fbc%2Fdocs%2Fsupport%2FUCR%2FSupportManual%2FTPM_aa-rtehc-te%2FTPM_aa-rtehc-te.pdf

Cheers
Peter

[Mike Naime]

Leonid:

The answer to your question depends on what you are trying to accomplish, and what you are trying to prevent.

We offer a DR solution to our clients where we host a Standby DB that is recieving Archive logs that are applied on the standby DB after a 24 hour delay. This is to prevent DB corruption of both DB's at the same time. If Oracle corruption occurs, we can roll our DB forward to just before the corruption and get the client back online in a couple of hours. This is much faster than a full DB restore and Archive log application would take the client to perform.

So to summarize, if you want to use the CA. Don't use the Archive logs. If you want to use the Archive logs. Don't use CA.


Have you tried a SNAP of the remote mirror drive? Presente the SNAPed drive in Read-only mode to your other cluster?

If you are wanting to perform backups. you can SNAP the drive. Perform your backup, and send the backup to tape. Then delete the snap. (Repeat tomorrow!)

[Orrin]

Hi,

By not having all the 4 nodes in the one cluster you defeat the principle behind tru Cluster 5.1

Additionally, keeping two servers idle when you have Oracle 9i RAC, doesn't make sense.

The only reasons that come to mind are if you can't afford the liscencing required for Oracle or the cluster connectivity requirements.

Anyway, you cannot use Oracle Archive log and CA together.

A better solution would be to create a SNAP Clone and present that to the other hosts. The prolem with this solution is the data will not be in sync at all times, and you will have to keep creating the snap clone at regular intervals.

A simpler solution would be to put all the nodes in a single cluster, set up a cluster alias for the primary oracle server and another for the oracle readonly. Then create services for the oracle primary and the oracle readonly. Have the caa daemon manage this acoording to your requirements.

Hope this helps,

Regards,
Orrin.

[Leonid Khrisanfov]

Thank you all. I know now that it's not possible to mount target device until failover. I do not have Business Copy to use snaps. But solution is found. The fact is that Customer has three dbases running. Two of them we're going to run at the first site and mirror to the second one, and third base will run at the second site and mirror to the first one. By this solution we will have performance of dbases and disaster tolerance.
Actually, I did it already, but new problem occurs. Contrary to my expectations degradation of system performance is observed.