PARALLEL LOAD - SPACE USAGE

Question

제품 : ORACLE SERVER
작성날짜 : 2004-08-13
SCOPE
8.1.6 이상의 Standard Edition 부터 지원이 되는 기능입니다.
Subject: Parallel Load - space usage.
I have put together a short summary of how Parallel Load handles
storage allocation in 7.1 which I am going to send to a customer.
This could also quickly turn into a bulletin or PR entry
(if one doesn't already exist).
I'd appreciate if people could review and give corrections/enhancements.
Here is how storage allocation works with parallel loads.
For each parallel loader session, Oracle
1) creates a temporary segment with the INITIAL and NEXT
from the table that is being loaded, and in the datafile
designated by the FILE option.
2) When a load is complete, the final extent used by that
parallel load is trimmed back to the last block which it
used.
3) Lastly, the extents belonging to the temporary segments
are then converted to extents belonging to the table
being loaded.
This creates a couple of problems:
* If your INITIAL and NEXT are different for the table,
then you will end up with multiple extents sized the
same as INITIAL, which doesn't follow general table
growth patterns.
* Extent sizes are not predictable for the last extent of
each load. This can lead to tablespace fragmentation.
What to do about this?
* Size INITIAL and NEXT to be the same.
This will eliminate the first problem listed.
* Oracle 7.2 will have a feature in to specify the storage
parameters used by the parallel load (bug 187319):
OPTIONS (STORAGE=(MINEXTENTS n
MAXEXTENTS n
INITIAL n K
NEXT n K
PCTINCREASE n))
* Enhancement Request 236552 has been filed to allow the
user to select no trimming of the extents.

Answer 1

Hi Karen,
The number of workprocess (Dialog/Background) could be reduced also by selecting the loading options in InfoPackage.
In InfoPackage Processing tab, there are various loading options,
PSA & Then into data target (Package by Package) - This option uses only one process at a time. One data package is lifted from the Source System by a background process to PSA, and then the same process lifts it from PSA to datatarget.
PSA & data target in parallel (Package by Package) - This option uses two process at a time. One data package is lifted from the Source System by a background process to PSA, and then another process lifts it from PSA to datatarget. The first process same time, gets busy in lifting another package from Source System to PSA.
The first option is good, but takes more time to load data than the second one, so there is a tradeoff.
Also it is good if you use another process in process chain to activate the ODS instead of setting it up on automatic activation.
Please let me know if you need any more help.
Regards,
Pankaj