Re: Parallel queries

  • From: "jaromir nemec" <jaromir@xxxxxxxxxxxx>
  • To: <greg@xxxxxxxxxxxxxxxxxx>, <kennaim@xxxxxxxxx>
  • Date: Sun, 27 Apr 2008 21:43:02 +0200

Hi Greg,

What is the logic of using 5 or 10 for DOP?
I would recommend an even number (powers of two, even better) for the
DOP.

Interesting recommendation, for sure valid while joining partition-wise with a hash partitioned table (but here the deriver is the number of hash partitions not the DOP).

In case of a non partition-wise hash join I can't imagine the first part (parallel scan of a table or partition) could profit of an even DOP. The DOP could have relevance in the distribution of rows between parallel slaves (especially for hash distribution) leading for "inappropriate" number of consumers to a skew distribution. Interestingly I neither heard some recommendation nor I read some case studies nor I had some bad experiences using inappropriate DOP for the hash distribution in parallel queries. (Even I had no fear to use a prime number DOP).
Am I missing something?

Regards,

Jaromir D.B. Nemec
--
//www.freelists.org/webpage/oracle-l


Other related posts: