Re: Informatica Bulk Mode behavior
- From: Tim Gorman <tim@xxxxxxxxxxxxx>
- To: <oracle-l@xxxxxxxxxxxxx>
- Date: Tue, 27 Apr 2004 09:00:28 -0600
Sounds like they're rationalizing the fact that they screwed up the
coding... :-)
They should be using "OCIDirPathDataSave()" for data saves, not
"OCIDirPathFinish()" followed by "OCIDirPathPrepare()". The amazing thing
is that they are aware of their screw-up! Their coding is analogous to
closing and re-opening a frequently-executed cursor after each execution --
unnecessary and foolish.
The fact that they coded incorrectly and that it is affecting space
management adversely, causing bulk loads to be perform unnecessarily poorly
as well, is bad enough. But the mis-information in blaming their faulty
coding on an "Oracle design limitation" is either dishonesty or stupidity or
both.
on 4/27/04 6:52 AM, Thomas Jeff at jeff.thomas@xxxxxxxxxxx wrote:
> Tim, Waleed, we got this response from Informatica Support:
>
> We are using the OCI DPL (direct path load) in the following way:
>
> First, We get a block with "x" # of rows.
>
> Do following loop
> Call ColArrayToStream for these "x" rows.
> Load whatever we can (call DirPathLoadStream)
> Do a DirPathStreamReset.
> while ColArrayToStream did not convert all the rows, repeat the above
> steps for the remainder rows.
>
> Every so often, depending on our "commit" interval, we will call
> DirPathFinish (and then reprepare).
>
> The 2 differences between our process and Oracle's demo program are:
> 1) the # of rows "x" may be different. They may be optimizing the "x"
> that they pick. We derive "x"=20
> from the buffer block size.
>
> 2) Oracle does 1 DirPathFinish at the end of the process, whereas we do
> it on every commit. =20
>
> Thus, we could probably simulate exactly what Oracle's test program is
> doing by making sure that the=20
> # of rows in the block matches their test program and the target commit
> interval is really high.
>
> So there seem to be 2 possible places where we may be causing more
> extents to be created than necessary. =20
>
> 1) The last load in the loop may be less than the amount of rows that
> can fit in a transfer block. Not sure if this matters at all, but if
> the last load contains only 1 row, this may be a wasted allocated
> extent.
>
> 2) The DirPathFinish DEFINITELY seems to be causing a new extent to be
> created.
>
> We still have one X factor that we need to account for. The CR mentions
> that if the commit interval is high, we=20
> sometimes get the following error:
>
> ORA-01658: unable to create INITIAL extent for segment in table space=20
> Our guess is that somehow we ran out of space to get an extent.
> (perhaps in the temp table space) Perhaps=20
> this can be limited somehow if there is a way to turn off logging (if it
> was, indeed, on), but I didn't see=20
> this option available. Although we haven't tested this yet, we would
> think that the sample program would have=20
> similar issues, although it may take a little longer because they may
> not have wasted extents as listed in option=20
> 1) above.
>
> Solutions:
> The workaround fix is to play with the buffer block size and the commit
> size. If the buffer block size fits=20
> the correct amount of rows, we may have a somewhat optimal fix. We may
> want to consider logging certain=20
> information that may help the user pick a correct buffer size, but, of
> course, this can be kinda cryptic.
----------------------------------------------------------------
Please see the official ORACLE-L FAQ: http://www.orafaq.com
----------------------------------------------------------------
To unsubscribe send email to: oracle-l-request@xxxxxxxxxxxxx
put 'unsubscribe' in the subject line.
--
Archives are at //www.freelists.org/archives/oracle-l/
FAQ is at //www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------
Other related posts:
