Re: 10g data pump question

From: joel garry <joel-garry_at_home.com>
Date: Fri, 21 Nov 2008 14:29:40 -0800 (PST)
Message-ID: <1b742839-0813-4b0d-8fd5-6e199cd64971@v39g2000pro.googlegroups.com>


On Nov 21, 9:16 am, Chuck <chuckh1958_nos..._at_gmail.com> wrote:
> DA Morgan wrote:
> > Chuck wrote:
> >> Why is it that when I run a data pump operation with PARALLEL > 1, I
> >> only ever see one worker EXECUTING and the state for all the others is
> >> WORK WAITING? What would cause that?
>
> >> TIA
>
> > My guess is that the reason is you have the syntax incorrect.
>
> > What likely you intended is:
>
> > parallel => 2
>
> > which is a named notation assignment of a value.
>
> My OP wasn't intended to show syntax, just that I've set PARALLEL to
> something greater than 1. The actual syntax is below (taken from a shell
> script)
>
> impdp / DIRECTORY=$DIRECTORY \
>   LOGFILE=$LOG_FILE_NAME \
>   NETWORK_LINK=IMPDP_${SOURCE_DB} \
>   SCHEMAS=${SOURCE_SCHEMA} \
>   REMAP_SCHEMA=${SOURCE_SCHEMA}:${TARGET_SCHEMA} \
>   PARALLEL=4
>
> Just seems like 3 of the 4 workers are pretty lazy and I wonder if
> there's any real benefit to setting parallel gt 1.

How many files do you have in the dump?

See metalink Note:365459.1

" The first worker begins to load all the metadata – the tablespaces, schemas, etc., until all the tables are created. Once the tables are created, the first worker starts loading data instead of metadata and the rest of the workers start loading data too.
Once the table data is loaded, the first worker returns to loading metadata again. The rest of the workers are idle until the first worker loads all the metadata up to package bodies. Multiple workers load package bodies in parallel. One worker loads metadata up to and including secondary tables. Multiple workers load secondary table data. One worker loads the remaining metadata.

Note: One worker creates all the indexes but uses PX processes up to the PARALLEL value so indexes get created faster.

Thus, an import job can be started with a PARALLEL = 10, and the user will only see one worker being utilized at certain points during job execution. No other workers or Parallel Execution Processes will be working until all the tables are created. When the tables are created, a burst of workers and possibly PX processes will execute in parallel until the data is loaded, then the worker processes will become idle. ...
For Data Pump Import, the PARALLEL parameter value should not be much larger than the number of files in the dump file set. "

jg

--
@home.com is bogus.
REBOOT
http://news.bbc.co.uk/2/hi/technology/7722367.stm
Received on Fri Nov 21 2008 - 16:29:40 CST

Original text of this message