Re: tunning an index build
- From: Fuad Arshad <fuadar@xxxxxxxxx>
- To: genegurevich@xxxxxxxxxxxx, Alexander Fatkulin <afatkulin@xxxxxxxxx>
- Date: Wed, 18 Jul 2007 09:40:50 -0700 (PDT)
if workarea_size_policy is manual then pga_aggregate_target is not used.
----- Original Message ----
From: "genegurevich@xxxxxxxxxxxx" <genegurevich@xxxxxxxxxxxx>
To: Alexander Fatkulin <afatkulin@xxxxxxxxx>
Cc: oracle-l@xxxxxxxxxxxxx
Sent: Wednesday, July 18, 2007 10:32:11 AM
Subject: Re: tunning an index build
Alex,
Yes, I have set pga_aggregate_target to 1.5G and workarea_size_policy is
manual. I thought that this will make
the sort_area_size (and all other area_size parameters) unused. IS that not
the case?
I am also working with the UNIX team to address the I/O contention that I
am seeing.
thank you for your input
Gene Gurevich
"Alexander
Fatkulin"
<afatkulin@gmail. To
com> genegurevich@xxxxxxxxxxxx
cc
07/17/2007 11:10 oracle-l@xxxxxxxxxxxxx
PM Subject
Re: tunning an index build
Do you use those 1.5G through pga_aggregate_traget?
If so, try to use more memory by switching your session to
workarea_size_policy=manual and dedicating as much to sort_area_size
as you can.
Make sure you spent enough time on temp I/O before doing that.
On 7/17/07, genegurevich@xxxxxxxxxxxx <genegurevich@xxxxxxxxxxxx> wrote:
> Hi all:
>
> I think I have posted on this issue before, but I can't find my old
Emails
> so - sorry for the repeated question.
> Here is the situation. I am running oracle 9208 and have a 40G table
with
> a lot of indices - mostly bitmap
> Their sizes are under 1G. These indices take about 20 to 30 minutes to
> build. That seems to be too long.
> I have just waited for 24 minutes to build a 82M index. The index is
> parallelized 4 way, the table is
> parallelized 6 ways. I have 2G sga and 1.5G pga. I do see a lot of IO
waits
> (via vmstat) and I have UNIX
> team looking into that. I think I have plenty of memory and the
parallelizm
> should be enough for a small index.
>
> Any thoughts on what else I should look at?
>
> thank you
>
> Gene Gurevich
>
>
> --
> //www.freelists.org/webpage/oracle-l
>
>
>
--
Alex Fatkulin,
The Pythian Group,
http://www.pythian.com/blogs/author/alexf/
--
//www.freelists.org/webpage/oracle-l
Other related posts:
