Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.server -> Re: Oracle 9.2 PGA_Aggregate_target and a statistics question
I have a note from somewhere (which might be the manual) that no process is allowed more than 5% of the total PGA_AGGREGATE_TARGET.
I believe that Oracle also has an algorithm for being parsimonious with memory if the P_A_T is under pressure.
Apart from that, it's still on my todo list.
Possibly oracle works out for each join the minimum memory required for 100% in-memory execution, the minimum memory required for a one-pass disc execution and the minimum for a two-pass execution and decides which it can afford. But that begs the question of handling a complex multi-layer hash, or a hash feeding a sort (group by) or sort (order by).
The biggest problem I find with the 10053 is that the more features Oracle gets, the less useful the 10053 becomes.
-- Jonathan Lewis http://www.jlcomp.demon.co.uk Next Seminars UK July / Sept Australia August Malaysia September USA (MI) November http://www.jlcomp.demon.co.uk/seminar.html Telemachus wrote in message ...Received on Thu Jul 11 2002 - 10:43:13 CDT
>PGA_AGGREGATE_TARGET : wonderful ! at last !
>
>
>... however I have a question ...
>
>Correct me if I'm wrong
>
>In the following PGA_AGGREGATE_TARGET = PAT
>1. premise :
>SORT_AREA_SIZE and HASH_AREA_SIZE are base inputs to the query optimization
>process so it can figure out how many passes etc etc, which table can fit
>and so on .
>
>2. premise :
>if these are not now directly specified , then it must use some fraction or
>the whole of PAT(minus of course whatever else is hogging PAT from everyone
>else's queries - for simplicity let's call the available bytes delta-PAT)
>
>questions :
>Is all of delta-PAT considered as the base input to both of these
parameters
>or does anyone know what the fraction is. ?
>My ev10053 reading skills have never been very good. The reason I ask is
>it's an interesting subtle change to the query optimization process, since
>now sort joins and hash joins have the potential to get a hell of a lot
more
>memory and we may come 'right down to the wire' finally for the final
>decision.