Re: Sort Area Size

In article <affbsf0171k_at_drn.newsguy.com>, Thomas Kyte <tkyte_at_oracle.com> wrote:
>In article <aff1pp$2t5$1_at_cronkite.cc.uga.edu>, jeff_at_work.com says...

>fine, tell me then, if
>
>sort_area_size = 1m
>sort_area_retained_size = 64k
>
>and you have MTS (shared server)
>
>a) how much UGA memory will be allocated at most for the sort
>b) how much PGA memory will be allocated at most for the sort
>c) will the contents of the PGA memory be MOVED to the UGA (since PGA is in the
>process and UGA is in the SGA)
>
>Answers
>
>a) 64k -- for the stuff that is retained after the sort. at most, could be 0 if
>this was an internal sort and didn't need to be fetched by the client.
>b) 1m -- at most, for the sort.
>
>c) yes, since the sort actually takes place in the PROCESS (shared server) but
>the results need to be in the SGA (for the client to fetch from)
>

So, basically, you're saying that the sort area is always allocated entirely in the PGA (never really deallocates from there, since the PGA size is supposed to remain fixed in size) and, after select sorts returning results to users, then copies itself over to the UGA, reallocating memory there according to the limit imposed by sort_area_retained_size?

>That is my point, nothing more nothing less. Yes, if differs from your
> reading,
>your interpretation but is valid none the less.

Again, *I* am not arguing with you, and it's not MY "interpretation." I've shown quite clearly where Oracle's own documentation and teaching materials are explicitly contradictory. I hereby Take Your Word For It, because I am perfectly aware that you know much more about this than I do. I'm just not at all happy being told one thing by Oracle and then finding out that it's all wrong... that it is "factually inaccurate mythology."

Thank you for your time and patience. Received on Thu Jun 27 2002 - 13:05:58 CDT