Transformation: moving large Oracle databases from AIX or Solaris to Linux with cross-platform transportable tablespaces

#Why XTTS

XTTS moves datafiles (converting endianness) and keeps them current with RMAN incremental backups until cut-over.

#How it works

1. Full backup of the tablespaces on source (online)  ──►  restore + convert on Linux target
2. Incremental backups (repeat)                       ──►  apply + convert on target
3. Cut-over: tablespaces READ ONLY, last incremental  ──►  apply
4. Data Pump metadata export (TTS)                    ──►  import metadata, plug tablespaces in
5. Make tablespaces READ WRITE on target, validate, switch apps

#Prerequisites

• Use Oracle's current XTTS V4 scripts from My Oracle Support (the "V4 Reduce Transportable Tablespace Downtime using Cross Platform Incremental Backup" note) and follow that note's version requirements.
• Target database created on Linux with the same (or compatible) character set, same block size(s), and the same or newer version.
• Tablespaces must be self-contained:

EXEC DBMS_TTS.TRANSPORT_SET_CHECK('APP_DATA,APP_INDEX', TRUE);
SELECT * FROM transport_set_violations;

• SYSTEM, SYSAUX, UNDO, TEMP are not transported. Users, roles, grants, PL/SQL in SYSTEM, sequences and public synonyms are moved separately (Data Pump with FULL metadata or targeted exports).
• Staging space reachable by both (NFS or copy).

#Configure and run (outline)

The V4 driver uses an xtt.properties file (tablespaces, source/target platform IDs, backup and restore directories, destination datafile location):

# source
$ORACLE_HOME/perl/bin/perl xttdriver.pl --backup     # first run: level 0 backup of the tablespaces
# copy files to target staging; on target:
$ORACLE_HOME/perl/bin/perl xttdriver.pl --restore    # restore + convert to Linux format

Repeat --backup / --restore as often as needed. Each run takes only the changes since the last one (block change tracking on the source makes this fast).

#Cut-over

1. Stop application, then on source: ALTER TABLESPACE app_data READ ONLY; (all transported tablespaces).
2. Final --backup / --restore.
3. Export metadata on source:

expdp system directory=DP_DIR dumpfile=tts_meta.dmp transport_tablespaces=APP_DATA,APP_INDEX transport_full_check=y logfile=tts_exp.log

4. Create users/roles on target, then import, pointing to the converted datafiles:

impdp system directory=DP_DIR dumpfile=tts_meta.dmp \
  transport_datafiles='/u02/oradata/NEW/app_data01.dbf','/u02/oradata/NEW/app_index01.dbf' logfile=tts_imp.log

5. Import remaining metadata (sequences, procedures, grants, synonyms) with a Data Pump CONTENT=METADATA_ONLY export of the schemas excluding tables.
6. ALTER TABLESPACE app_data READ WRITE;, gather statistics, validate and switch applications.

Full transportable export/import (FULL=Y TRANSPORTABLE=ALWAYS) can combine steps 3–5 for 12c+ sources. Check whether it fits your case.

#Validation

• Object counts and invalid objects by schema, source vs target.
• Row counts on key tables (tablespaces were read-only, so counts must match exactly).
• RMAN VALIDATE on the new datafiles.
• Application smoke tests, batch jobs, DB links.

Warning

Rehearse the whole cut-over on a copy first. The metadata import step and missing non-transported objects (sequences, grants on SYS objects, public synonyms) cause most cut-over surprises, not the datafiles.

#Checklist

• Latest XTTS V4 scripts and MOS note followed
• Self-contained tablespace check clean
• Block change tracking on source
• Incremental cycles timed; final incremental size known
• Non-transported objects scripted
• Full rehearsal with timings and rollback (source stays intact until switch)