Re: Minimize Performance Hit on Sort...Help!
Date: Tue, 2 Feb 2010 12:20:32 -0800
Message-ID: <a9c093441002021220q59b4e445j3f36d06d14df2060_at_mail.gmail.com>
I would comment that sorting large amounts of data parallel IS a good thing - that is the whole premise behind parallel computing. It's just that the sort operators have to be well placed. If your parallel sort is not performing, then the obvious question would be why (what does the ASH/AWR data show it is waiting on?)
When you say it "comes back" does that mean starts to return rows, or it completes? If a sort (order by) is requested, then no rows can be returned to the client until they are ordered, where as w/o an order by they can be returned immediately.
Can you post the execution plans for the order by and w/o order by?
On Tue, Feb 2, 2010 at 8:57 AM, Kellyn Pedersen <kjped1313_at_yahoo.com> wrote:
>
> I'm at a total brain block on this one- Query with a order by that MUST be done in Oracle as the file is an output to a dataset in SAS. The query is through a view and I would prefer to avoid parallel, as this makes the sorting problem worse, (sorting large amounts in parallel is just NOT a good thing...)
>
> The simple query with the order by-, (without the order by, the data will come back in seconds...)
>
> select smap.* from CENSUS_0_48 smap order by ibhid, ibid;
-- Regards, Greg Rahn http://structureddata.org -- http://www.freelists.org/webpage/oracle-lReceived on Tue Feb 02 2010 - 14:20:32 CST