Re: Data Migration options - oracle to oracle

From: rjamya <>
Date: Fri, 6 Sep 2013 11:34:49 -0400
Message-ID: <>

I think the order comes from total size of the table as computed in the 'estimation' phase. with Parallel, one thread does metadata, and then while it does that, master process will dish out one table at a time via internal queues to each of the available worker threads. again in the order of total size/rows. One the thread that does metadata completes metadata, it takes next available table and starts processing. in non-parallel mode this is all sequential as seen from logfile.
This is just my observation but if you are curious, hopefully you can easily verify this by running trace, I believe this allocation is done by master process (just a wild guess), I have done tracing only once so far. One of these days I'll verify if it does order by blocks in the table or rows (or some secret algorithm in between).


On Fri, Sep 6, 2013 at 11:23 AM, Norman Dunbar <>wrote:

> With parallel=2 I saw the nice ordered sequence of "processing object
> ...." followed by the "exporting ....." messages, interspersed. Lots of
> "exporting..." and a few "processing..." messages all randomly mixed up.
> So, I stand corrected on the alphabetic point I made, thank you very much.
> As for the order it decides? Who knows. Maybe it's related to how the
> objects names come back in a "select table_name from user_tables" perhaps?
> (No, it's not - I tried that too!)

Received on Fri Sep 06 2013 - 17:34:49 CEST

Original text of this message