Re: Why does Oracle sort hashjoin output before parallel insert?

On Sat, 6 Mar 2004 18:18:18 +0000 (UTC), "Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote:

>
>Just one more thought.
>
>If you get a parallel insert, then the distribution
>after the hash join will be a range-based distribution,
>so perhaps the intermediate result set (between the
>hash processes and the insert processes) accumulates
>in a temporary segment that is labelled as a sort segment
>because some sorting has gone into producing the
>payload for each insert slave.

I tried to monitor the refresh to discover if any real sorting was going on, but the monitor script aborted along the way, before I could check your assumption.

In the night from tuesday to wednesday I will be able to do a refresh with parallel dml disabled. If I'm not mistaken this should be much quicker because joined records are written to table immediately, although things should slow down a bit because there is only one process doing direct load inserts. On thursday I come back with the results.

Jaap. Received on Mon Mar 08 2004 - 12:18:45 CST