Re: expdp is painfully slow
Date: Wed, 7 Apr 2010 09:02:36 -0700 (PDT)
On Apr 6, 5:33 pm, zigzagdna <zigzag..._at_yahoo.com> wrote:
> On Apr 6, 2:53 am, DG problem <skatef..._at_gmail.com> wrote:
> > On Apr 5, 5:01 am, zigzagdna <zigzag..._at_yahoo.com> wrote:
> > > I am using Oracle 18.104.22.168.1 on HP UNIX 11.0. I ma using expdp and
> > > find it extremely slow. However. Most of the waiting time is during
> > > metadata (see below) where it waits for several hours. When it starts
> > > exporting the rows, it progresses reasonably fast.
> > > Processing object type DATABASE_EXPORT/SCHEMA/TABLE/INDEX/DOMAIN_INDEX/
> > > INDEX
> > > I have seen some threads on Metalink as well as on Google. All otf
> > > tem point to these problems in Oracle 10.2 expdp version, but I see
> > > same problem in Oracle 22.214.171.124.1 as well. A suggestion to exclude
> > > STAITISTICS has not helped me:
> > > expdp system/password directory=expdp_dir dumpfile=$BackupExportFile
> > > logfile=$TLOGFILE \
> > > full=Y compression=all exclude=STATISTICS TRACE=480300
> > > I have turned on expdp tracing and I see statements like following for
> > > various indexes and do not give me any clue to speed things up.
> > > KUPW:14:48:25.668: 1: Base Process info: 6122 and processing status: C
> > > and processing state: R
> > > KUPW:14:48:25.668: 1: new state: R
> > > KUPW:14:48:25.668: 1: new status: C
> > > KUPW:14:48:25.668: 1: In procedure BUILD_SUBNAME_LIST with
> > > INDEX:MAXDEMO71.CONTRACT_NDX2
> > > KUPW:14:48:25.669: 1: In function NEXT_PO_NUMBER
> > > KUPW:14:48:25.669: 1: FORALL called.
> > > KUPW:14:48:25.676: 1: FORALL returned.
> > > KUPW:14:48:25.678: 1: DBMS_LOB.TRIM called. v_md_xml_clob
> > > KUPW:14:48:25.679: 1: DBMS_LOB.TRIM returned.
> > > KUPW:14:48:25.679: 1: DBMS_METADATA.FETCH_XML_CLOB called. Handle:
> > > 200001
> > > Is ether anyway to speed expdp.
> > About 3 to 4 years ago I experienced the same problem with 10g. I
> > narrowed down the problem to the export of the users. I had about
> > 120,000 users in the DBA_USERS table (PeopleSoft database) and so
> > during testing at one stage I dropped all the users and ran expdp
> > again and it ran much quicker.
> > Anyway, that was from memory so I hope I have recalled it properly.- Hide quoted text -
> > - Show quoted text -
> I have narrowed down the problem to domain indexes which get created
> because of Oracle text indexes which my application needs.
> Unfortunately, I cannot get rid of domain indexes so stuck with it god
> knows until when.
> I have yet another problem with 11g, cannot create db control most of
> the time. Oracle support looked at it and there is a bug about it on
> metalink, yet they will not document my problem as a bug and fix it.
I have a 10g call open (several issues, Itanium hp-ux 11.23), seems to be difficult because it really is a different product than the db, and the whole EM support seems based on giving it its own home environment. So try setting up an environment specific for dbcontrol, and only start, stop, rebuild, control or modify it in that environment, and have the same environment for system start and online (if you have automatic startups). In the 10g case at least, it seems important to use the built-in version of java, rather than the "or greater" version the docs say (I'm still questioning support about this assertion, though it makes sense in this context). In other words, don't have any references to OS versions of perl or java in your path, with oracle_home first. I'd be curious as to what perl or java advocates have to say about that.
-- _at_home.com is bogus. hactivism, lose tenure, DoS, TIT, ivory tower, academic freedom: http://www.signonsandiego.com/news/2010/apr/06/activist-ucsd-professor-facing-unusual-scrutiny/Received on Wed Apr 07 2010 - 11:02:36 CDT