Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: cache vs keep buffer pool

Re: cache vs keep buffer pool

From: Daud <daud11_at_hotmail.com>
Date: 26 Aug 2002 22:39:50 -0700
Message-ID: <f0bf3cc3.0208262139.34d21e1@posting.google.com>


"John Beresniewicz" <jberesniewicz_at_comcast.net> wrote in message news:<Rhxa9.212517$m91.8802236_at_bin5.nnrp.aus1.giganews.com>...
> "Daud" <daud11_at_hotmail.com> wrote in message
> news:f0bf3cc3.0208260711.5490ca76_at_posting.google.com...
> > If I have a small table that is accessed very frequently what is the
> > advantage of moving it over to the keep buffer pool as opposed to just
> > 'cache' it using the cache storage parameter? If it is accessed very
> > frequently the blocks should be cached in the buffer pool and chances
> > of them getting aged out should be slim.
> > So, what advantage would the keep buffer pool offer in this case?
> >
> > Another question.
> > With multiple db writers and db block lru latches, how do the db
> > writers get assigned to do all the work? The documentation says that
> > they are assigned to the lru latches (working sets) in a round robin
> > fashion. If that is the case, how is it possible for the db writers to
> > be NOT equally loaded?
> > Maybe I am missing some points here. Am I?
> >
> > rgds
> > Daud
>
> As far as DBWR loading is concerned, consider the following setup:
>
> 2 buffer pools, DEFAULT and KEEP
> 3 LRU latches assigned as follows: 2 to DEFAULT, 1 to KEEP
> 2 DBWR processes
>
> Latches are assigned SEQUENTIALLY to pools and ROUND ROBIN to DBWRs:
>
> LRU1 => DEFAULT => DBWR1
> LRU2 => DEFAULT => DBWR2
> LRU3 => KEEP => DBWR1
>
> Assume each working set within a pool presents essentially identical
> workload, so the two DBWR processes do the same amount of work from the
> DEFAULT pool. However, DBWR1 additionally is handling all the work from the
> KEEP pool, hence the imbalance.
>
> Avoid the potential for this imbalance by always assigning a number of LRU
> latches to each pool that is some multiple of the number of DBWRs.
>
> HTH
>
> ps - In Oracle 9i don't assign latches explicitly and allow the system to
> default based on the number of DBWRs.

From your example, does it mean that after the assignment, LRU3 will always be handled by DBWR1 only...never by DBWR2? If that's the case, it is not really round robin (well, only during the initial assignment).

rgds
Daud Received on Tue Aug 27 2002 - 00:39:50 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US