Migrating Oracle Database from Exadata to EMC

I would use ASM mirroring, with one side of the mirror being on ExaData and the other side being on EMC SAN. Then break the mirror. Fairly simple, really.

Excellent reply Jeff, and which Protocol would you use at the SAN level and which interface on the Exadata would you recommend connecting to?

Exadata storage consists of 14 federated servers( Full rack). Any data move from/to the Exadata requires host based movement. A SAN based solution is out of the question.

The are many approaches but it depends on your technical and business requirements.

An elegant solution would be to create a standby database  on a VNX or VMAX.

You can leverage Replication Manager and SnapView to create a clone of the EMC storage for HA/DR or populate test/dev. I would ensure there are no table in HCC on the Exadata.

Once everything checks out, you can make the target the primary and your HA/DR solution would also be in place.

Good Luck.

Randy

Without adding hardware, which would invalidate support, I believe that you can use iSCSI or NFS over the Ethernet or IB interfaces to present external storage to Exadata.  There are plenty of  posts  out there on the Oracle related forums discussing this.

If iSCSI was proposed as an “ASM mirroring” solution, it would involve rebalancing at least one existing Failure Group by adding iSCSI  LUNs and dropping Exadata disks. With HCC enabled and users still accessing the database, I would expect that Exadata would raise a storage error: ORA-64307

As the various “patch madness” posts floating around the internet and twitter indicate, there are always plenty of Exadata and ASM bugs out there. Here are three recent bugs from a normal MyOracleSupport  search on “Exadata ASM”

Bug 13842894 : ASM REBALANCE OF DROPPED DISK FAILED TO COMPLETE 2 WEEKS BEFORE MIRROR FAILED. (04/12)

Bug 14115120 : DSKM PROCESS APPEARS TO BE HUNG. INITIATING SYSTEM STATE DUMP (05/12)  

Bug 14646534 : MULTIPLE DRIVES MISSING FROM ASM - BOTH PARTNERS ON MIRROR SET (09/2012)

It would be a braver man than myself who would propose this as a viable Exadata exit strategy.

If the point of the solution is to get the database off of Exadata with the minimum risk and minimal loss of service. Would this achieve it?

Whilst ASM rebalance may be a technically feasible solution, Oracle Data Guard sitting on an EMC array would be  a simpler, less disruptive and much lower risk  exit strategy?

This whitepaper was produced by the Strategic Solutions Engineering team as a method for protecting Exadata with Oracle Data Guard on EMC VNX. It shows how an Oracle database sitting on Exadata can be protected, backed up and re-provisioned elsewhere by combining Oracle and EMC technologies. It also shows how to deal with databases where HCC is used.

Another option, if you cannot tolerate downtime, would be to use database replication tools. Oracle's Golden Gate would be an option but I expect Oracle not to be very helpful with support and offering a reasonable price to use this tooling for a limited period for migration only.

EMC partners with Informatica and they have a comparable tool (Informatica Data Replication): http://www.informatica.com/us/products/data-replication/data-replication/

If I'm not mistaken this is the software formerly known as WisdomForce (Informatica aquired them).

This tooling would allow users to replicate (online, bi-directional) between different Oracle versions or even (one-directional?) from Oracle to non-Oracle (DB2, Postgress, SQL-Server, MySQL, ...)

Quest Shareplex (now part of Dell) is another potential tool offering similar features.

Bart

These are all valid options but the add cost and developers complexity into the mix, eg Golden Gate is $17.5k per 2 cores on x86 allowing for the Oracle core factor core processor licensing factor + support $3.85k.

Oracle Data Guard sitting on EMC infrastructure is not the only possible solution but I think it is a good one for EMC customers in terms of cost/benefit, downtime and risk.

If HCC is not used and Dataguard is run in maximum protection the switchover time is minimized. If a customer cannot tolerate any downtime, it would not be unreasonable to expect them to have designed failover into their application/infrastructure. Oracle provide a very good guide on this subject.

http://www.oracle.com/technetwork/database/features/availability/maa-wp-11gr2-client-failover-173305.pdf

Allan

Allan,

Good points. I am aware that Oracle charges the max for Golden Gate, that's why I pointed out that there are similar tools on the market that typically are less expensive.

Data Guard is an option and can be free of license (if you don't use Active Data Guard), however, it requires the exact same database version and structure on the target, which limits you to same-platform-same-version migrations. No way to get from Big-endian to little-endian, for example. GG and Informatica can migrate from, say, Oracle 9i to 11g, without application downtime, and you can run mixed mode (i.e. performing transactions against the old and new database simultaneously because it is bi-directional). Going live then is nothing more than disabling the old environment.

Also, DG does not allow for HCC-to-non-HCC replication. So if you have HCC then you have to decompress HCC afterwards (downtime). With GG or Infa Data Replicator, you might be able to convert on-the-fly.

But yes, there's no such thing as free lunch :-P

Cheers

Bart