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: On AIX, blocksize an DataWarehouses

Re: On AIX, blocksize an DataWarehouses

From: Frank R. <trek1s_at_gmail.com>
Date: 21 Nov 2006 11:01:18 -0800
Message-ID: <1164135678.531774.168340@k70g2000cwa.googlegroups.com>

On 21 nov, 17:29, DA Morgan <damor..._at_psoug.org> wrote:
> Frank R. wrote:
> > Hello everybody,
>
> > I'm currently involved in a Siebel Analytics/Datawarehouse deployment.
> > The datawarehouse database (9.2.0.8 AIX 5.3) was originally created
> > with a 8kb blocksize (standard) which as far as i've read anywhere or
> > asked anyone, is quite small, reccommending 32kb instead.
>
> > Apart from a stripping strategy between several disks, blocksize seems
> > critical in I/O improvement and I need advice on the relationship
> > between oracle blocksize and AIX blocksize.http://www.dizwell.com/prod/node/58
>

Thanks for the link. Very interesting, really.

> > ATM, the OS blocksize is
> > 4kb but, what would be the perfect size for it ¿32k too? ¿any oracle
> > blocksize multiple?
>
> > Thanks in advance for your help.Based on what metric(s) are you concluding that block size has any
> relationship to any performance issues you are trying to correct?
> If you don't have numbers that support the conclusion you are most
> likely going in the wrong direction.

Yeah, I know, i need "much" more testing, but:

Top 5 Timed Events



% Total
Event                                               Waits    Time (s)
Ela Time
-------------------------------------------- ------------ -----------
--------
PX Deq Credit: send blkd                           98,776      13,100
 51.44
db file sequential read                           341,578       6,892
 27.06
PX Deq Credit: need buffer                         53,359       3,447
 13.53
buffer busy waits                                  23,228         968
  3.80
log buffer space                                    1,012         333
  1.31

Also, 600 ms datafile read access seemed bad so, right now the data is being redistributed between all disks

Moreover, I'm using async I/O, gonna try direct I/O (where db_block_size "doesn't matter") instead, and see the differences.

Thanks again.

> --
> Daniel A. Morgan
> University of Washington
> damor..._at_x.washington.edu
> (replace x with u to respond)
> Puget Sound Oracle Users Groupwww.psoug.org
Received on Tue Nov 21 2006 - 13:01:18 CST

Original text of this message

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