Re: Question - any parameter or hint to FORCE Physical Read (bypass buffer cache)

From: Alex Fatkulin <afatkulin_at_gmail.com>
Date: Mon, 22 Apr 2013 16:57:20 -0400
Message-ID: <CAMVw97LGbVbM_tHG6fzf8vzFUGG93hR4WWSNb9-p9-KYHwDOkQ_at_mail.gmail.com>

Working with "_serial_direct_read"=true is harder because if direct path reads are not happening you will be left guessing whether it's the operation in the plan which doesn't support it or some of the variables used to determine whether serial direct path reads should happen or not are not adding up.
What you can do is:

1. Run your full table scan in parallel, PQ slaves will do the FTS via direct path reads since there is no in-memory PQ in 10.2
2. Allocate a small keep/recycle pool (I assume you don't want to set the default cache too low for the same reasons you don't want to flush it) and put the table in there.

On Mon, Apr 22, 2013 at 4:49 PM, <Christopher.Taylor2_at_parallon.net> wrote:

> Forgot a piece – Yes I knew about the FULL requirements on table/index
> scans and the query is using a simple full tablescan.****
>
> ** **
>
> Chris****
>
> ** **
>
> *From:* Taylor Christopher - Nashville
> *Sent:* Monday, April 22, 2013 3:48 PM
> *To:* 'Tanel Poder'; Alex Fatkulin
> *Cc:* ORACLE-L
> *Subject:* RE: Question - any parameter or hint to FORCE Physical Read
> (bypass buffer cache)****
>
> ** **
>
> The amount of IOs per execution I haven't checked. I should/will check
> that.****
>
> ** **
>
> The plans for the query I'm interested in are really simple and are the
> same on all 3 nodes.****
>
> ** **
>
> Chris****
>

-- 
Alex Fatkulin,
http://afatkulin.blogspot.com

Enkitec,
http://www.enkitec.com

--
http://www.freelists.org/webpage/oracle-l

Received on Mon Apr 22 2013 - 22:57:20 CEST