Because you never said it wasn't a sort end that wasn't releasing the
emory - by default one would assume it was a session end. Start at the easy
bit !
Get the Windows Server Resource Kit to look at memory swap etc by process.
"Chuck" <chuckh_at_softhome.net> wrote in message
news:Xns932CA804D7B36chuckhsofthomenet_at_130.133.1.4...
> "Telemachus" <telemachus_at_ulysseswillreturn.net> wrote in
> news:Cyr6a.11541$V6.15610_at_news.indigo.ie:
>
> > Someone (can't remember who - think it was Mr. Kyte) put together a
> > lovely demo of PGA allocation (and the importance of
> > SORT_AREA_RETAINED_SIZE) that proves what's going on.
> >
> >
> > However - if I use SQL*Plus on W2K on the database server(dedicated
> > server - not MTS) , do a big sort/join (Sort_area_size, _retained_,
> > and hash_Area_size are all big too ) then in the taskbar we can see
> > Oracle taking a lot of memory. When I quit out of SQL*Plus or
> > otherwise end the session we can see Oracle release all that PGA
> > memory.
>
> Why do you have to disconnect to release the memory? According to the 8i
> manual, the memory is supposed to be released down to
> sort_area_retained_size whent he sort phase is finished. When all of the
> rows are returned to the app, the rest of the memory is supposed to be
> released. Our app gets to the point, but it does not appear that the
> memory ever gets released. BTW most of our users never disconnect from
> the database (and if the docs are right, they shouldn't have to).
>
> According to the metalink docs on NT/2k memory architecture (which btw I
> have read Sybrand), it is the VirtualAlloc set of function calls that's
> used to dynamically allocate/deallocate memory for session memory, so the
> memory should be returned to the o/s right away.
>
> I will loop for the Tom Kyte article you mentioned as well as the Ixora
> article.
Received on Tue Feb 25 2003 - 03:50:08 CST
