EXPDP is too slow even though the value of cursor_sharing changed to EXACT.

Question

Hi
We are having a 10g standarad edition database (10.2.0.4) on Solaris 5 which is RAC with ASM. Infact we are planning to migrate it to LINUX x86-64 and to 11.2.0.3. The database size is around 1.3 TB. We are planning to go with an expdp backup and impdp to new server and new version database.
SQL> select * from v$version;
BANNER
Oracle Database 10g Release 10.2.0.4.0 - Production
PL/SQL Release 10.2.0.4.0 - Production
CORE 10.2.0.4.0 Production
TNS for Solaris: Version 10.2.0.4.0 - Production
NLSRTL Version 10.2.0.4.0 - Production
SQL> !uname -a
SunOS ibmxn920 5.10 Generic_127128-11 i86pc i386 i86pc
As per the plan I started the expdp. But unfortunately the processing of tables even continued for one and half days and the backup didnt started also. After going through few docs I found that the CURSOR_SHARING should be EXACT to make the expdp more faster (Previously it was SIMILAR). So I changed the parameter to EXACT in one of the node and started the backup again yesterday night on the same node where I change the parameter. When today I came back still the processing going on. I checked the job status and found that the table processing is still going. It is not hanged at all. But its too slow.
What could be the reason. Here is the memory details and kernal parameter details.
Mem
Memory: 24G phys mem, 6914M free mem, 31G swap, 31G free swap
Kernal Parameters
forceload: sys/msgsys
forceload: sys/semsys
forceload: sys/shmsys
set noexec_user_stack=1
set msgsys:msginfo_msgmax=65535
set msgsys:msginfo_msgmnb=65535
set msgsys:msginfo_msgmni=2560
set msgsys:msginfo_msgtql=2560
set semsys:seminfo_semmni=3072
set semsys:seminfo_semmns=6452
set semsys:seminfo_semmnu=3072
set semsys:seminfo_semume=240
set semsys:seminfo_semopm=100
set semsys:seminfo_semmsl=1500
set semsys:seminfo_semvmx=327670
set shmsys:shminfo_shmmax=4294967295
set shmsys:shminfo_shmmin=268435456
set shmsys:shminfo_shmmni=4096
set shmsys:shminfo_shmseg=1024
set noexec_user_stack = 1
set noexec_user_stack_log = 1
#Non-administrative users cannot change file ownership.
rstchown=1
Do I need to make changes above of these. The dump is taking to local file system.

Answer 1

Hi,
I'd be looking at doing this in parallel over a database link and completely miss out sending anything to nfs - it will make the whole process quicker (you effectively skip the export part and everything is an import into the new instance).
I ran a 600GB impdp in this way over a db link and it maybe took 12 hours (can't remember exactly) - a lot of that time is index build in the new database so make sure your pga etc is set up correctky for that.
LOB data massively slows down datapump so that could be the issue here also. You should be able to acheive the whole process in less than a day (if you have no lobs...)
Cheers,
Harry