Need advice migrating from AIX 7 filesystems to Exadata Linux ASM - Large DBMS

Question

We are using 11.2.0.3 and 11.2.04 databases on AIX 7.1 using AIX filesystems .   We have some 2TB databases and some much smaller.  About 50 production and 200 non-production databases.   We are migrating to Exadata 4 with Linux .   What is your advice on the method of migrations that should be used.   We may be able to take some outage time to do this.  

Answer 1

I echo the data pump export/import recommendations. I've used data pump several times to migrate databases to Exadata - including an environment with a few DBs on AIX Power PC to Exadata last year. If you can take downtime, it is the simplest, most flexible and least risky method - and if you can put a little thought and extra effort into it can still be very performant. On Exadata it's good to setup the environment according to Oracle's published best practices - which usually means some configuration changes from your source. Data Pump allows you to set this up first and have it ready to go - then do the migration into a properly configured database. You can also put the source DB into read-only while the migration takes place if that helps the downtime requirements.
Some suggestions to maximize performance and limit downtime:
Consider using DBFS file system on the Exadata, and then mount it using NFS to your source DB servers, for the data pump file location. This may take a little longer on the export, but avoids having to do a separate copy of the files over the network afterward and can make up the time. Once on Exadata, importing off the local DBFS can really perform well.
Use parallelism with data pump to speed up the export and import. The degree will need to be determined based on your CPU capacity, but parallelism will speed up the migration dramatically.
If you're licensed for compression - use the compression with Data pump to minimize the file size.
Precreate all your tablespaces first, and possibly even the schemas - this goes back to setting things up according to Exadata best practices. You can potentially use HCC and other things on the Exadata tablespaces if you so choose. You can always use the data pump mapping if you want to change a few things about the tablespace names and such from the source.
If you're really trying to maximize the performance and minimize downtime, you can spend some time pulling out the DDL for your indexes and constraints from the source - and have them scripted. Then only export the data, not the indexes and constraints, and after the data is imported use your DDL scripts, with high degrees of parallelism, to create indexes and constraints afterward. Don't forget to alter the index objects to remove the parallelism afterward so not to leave a bunch of high parallel indexes in place. This method can usually perform much faster than letting data pump do this.
Test well, and look for objects that don't migrate correctly or well with data pump and potentially use SQL scripts to bring them over manually.
Look for opportunities with some objects, for example meta data or DDL that doesn't change, to pre-create on Exadata before taking the downtime and starting the migration.
HTH,
Kasey