I recently moved a 2TB database from Linux, Oracle 11.1 to AIX, Oracle 11.2 using impdp with network_link and parallel options. The data was imported in about two hours and the indexes took another 7 hours. There was no downtime but we kept the users off the system while this was happening.
Here is the command I used on the new server:
nohup impdp system/secret NETWORK_LINK=olddb FULL=y PARALLEL=25 &
Then you can run this to see which parallel process is importing which table:
impdp system attach Import> status Import> parallel=30 << this will increase the parallel processes if you want
I had created all the datafiles in the new database and created a db link named olddb beforehand. The servers were on the same local area network and the network was the bottleneck.
Steve Harville's Blog 
Pingback: Migrating a 2TB database to a new operating system with different endianness – All Things Oracle
Why did you use nohup? It was usefull using the exp/imp utility but wiht impdp there is no need anymore to do that.
LikeLike
Hi Vil,
Thank you for your comment. I may be a little paranoid 🙂 Here’s the Oracle documentation on using the interactive client: http://docs.oracle.com/cd/E11882_01/server.112/e22490/dp_import.htm#autoId64
LikeLike
Hi, Steve
How could you do that, could you mind to introduce your hardware environment ? Thanks.
LikeLike
The target hardware was a IBM P-Series virtual system with 48GB memory, (8) 6400MHz CPUs, (4) virtual Fibre Channel adapters, (1) 10Gb Ethernet port, with ASM on EMC Symmetrix VMAX. The source hardware was an Intel 32 core system with 16GB memory.
LikeLike
Reblogged this on MY DBA Notes.
LikeLike
Thanks for this overview…I’m trying to migrate from aix to Windows. I noticed that you created all the datafiles in the new database. Does the import map the source datafiles to the locations you created? In my case, would I need to do this also?
LikeLike
Hi Michael, Yes, pre-create the tablespaces. You can check the current sizes and size the new ones based on actual usage possibly saving some space on the target system.
LikeLike
To speed it up further, you can exclude indexes, constraints and statistics. Export the DDL for the indexes and constraints. Create the constraints with enable novalidate and indexes with parallel. Then set indexes to noparallel (or original degree) after building. Finally gather statistics. You can choose to validate the constraints later if you like.
LikeLike