From: Rick Denoire on
Hello

After Upgrading a commercial application from 9.2.0.5 to 10.2.0.3 we
got a disastrous performance, even though the new server was at least
10 times faster (in terms of CPU). The application manager did not
follow test procedures prior to the upgrade, which lead us to a
crisis.

After some investigation I found that just setting the following
parameter resulted in a huge (about 100-fold) performance increase:
OPTIMIZER_SECURE_VIEW_MERGING=FALSE

See for example:
http://www.freelists.org/archives/oracle-l/07-2006/msg00260.html

I did not dare to set any hidden parameters yet; perhaps those would
have boosted performance even more.

Can you confirm that reverting mentioned parameter to the non-default
value (FALSE) improves performance so dramatically?

What is the rationale of such detrimental default value of the
initalization parameter that effectively renders the database
"misconfigured"?

Are there any other deadly preset parameters I should be aware of
after Upgrade to Oracle 10g, according to your experience?

Do you really recommend to let Oracle do things automatically? (SGA,
PGA, db_file_multibloc_read_count, etc.)

Thanks a lot

Rick Denoire
From: Robert Klemme on
On 02.12.2007 02:30, Rick Denoire wrote:
> Hello
>
> After Upgrading a commercial application from 9.2.0.5 to 10.2.0.3 we
> got a disastrous performance, even though the new server was at least
> 10 times faster (in terms of CPU). The application manager did not
> follow test procedures prior to the upgrade, which lead us to a
> crisis.
>
> After some investigation I found that just setting the following
> parameter resulted in a huge (about 100-fold) performance increase:
> OPTIMIZER_SECURE_VIEW_MERGING=FALSE
>
> See for example:
> http://www.freelists.org/archives/oracle-l/07-2006/msg00260.html
>
> I did not dare to set any hidden parameters yet; perhaps those would
> have boosted performance even more.
>
> Can you confirm that reverting mentioned parameter to the non-default
> value (FALSE) improves performance so dramatically?
>
> What is the rationale of such detrimental default value of the
> initalization parameter that effectively renders the database
> "misconfigured"?
>
> Are there any other deadly preset parameters I should be aware of
> after Upgrade to Oracle 10g, according to your experience?
>
> Do you really recommend to let Oracle do things automatically? (SGA,
> PGA, db_file_multibloc_read_count, etc.)

I guess with this information given you cannot expect profound help.
What did you do to pinpoint the source of the slowness? I mean, do you
*know* where time is spent? If not, I'd investigate this first before
doing any changes to the system.

Kind regards

robert
From: Noons on
On Dec 2, 12:30 pm, Rick Denoire <educacion.super...(a)online.de> wrote:


> Can you confirm that reverting mentioned parameter to the non-default
> value (FALSE) improves performance so dramatically?

in some cases it might, depends on what the
views are like in your application.
If it works, then use it!


> What is the rationale of such detrimental default value of the
> initalization parameter that effectively renders the database
> "misconfigured"?

not really. it's very dependent on the type of
views and the application itself. I've seen
this not take any effect whatsoever and I've
seen it make a huge difference. Go figure...


> Are there any other deadly preset parameters I should be aware of
> after Upgrade to Oracle 10g, according to your experience?

_unnest_subquery=true (default)

even in 10.2.0.3 patched, I'm still getting
a lot of complex Peoplesoft queries with
problems with this one, particularly with the
ubiquitous (select max(date_field) ...)
subqueries of this product. Set it to false
and problems vanish.


> Do you really recommend to let Oracle do things automatically? (SGA,
> PGA, db_file_multibloc_read_count, etc.)

if you don't have multiple schemas then it's fine.
If you do, then watch out for tables with the
same name in different schemas and auto sga:
deadly combination!
From: csn on

> After some investigation I found that just setting the following
> parameter resulted in a huge (about 100-fold) performance increase:
> OPTIMIZER_SECURE_VIEW_MERGING=FALSE

If I can improve performance 100-fold, I wouldn't worry about other
things too much but go to the manager and ask for a pay raise instead.
From: Rick Denoire on
Hello Robert

I spent two weeks (!) investigating the question.
I required support from the vendor, sent them detailed informations.
I let users do different tests.
I collected and printed performance numbers, execution plans, looked
into waits, used AWR, ADDM, SQL Tuning Advisors, statistics
with/without histograms. I was playing with different init parameters
(mostly making things even worse). I set up a clone database, did more
tests... hey! Let's stop.

As you can see from other replies, you are guessing wrong. (Typical
case when people try to complicate obvious things).

If the software is just misbehaving, all this knowledge is for
nothing. Almost every query was doing FTS of huge tables, main
activity was direct path read. Same application was still installed in
the old server running 9i: Execution plans were completely different.
So what? It could have been something different than direct path
reads, it does not bear any direct logical relationship to the
solution. Based on this "knowledge", you would probably... buy more
disks?? Wrong!

Thanks anyway.

Rick Denoire


Robert Klemme <shortcutter(a)googlemail.com> wrote:

>On 02.12.2007 02:30, Rick Denoire wrote:
>> Hello
>>
>> After Upgrading a commercial application from 9.2.0.5 to 10.2.0.3 we
>> got a disastrous performance, even though the new server was at least
>> 10 times faster (in terms of CPU). The application manager did not
>> follow test procedures prior to the upgrade, which lead us to a
>> crisis.
>>
>> After some investigation I found that just setting the following
>> parameter resulted in a huge (about 100-fold) performance increase:
>> OPTIMIZER_SECURE_VIEW_MERGING=FALSE
>>
>> See for example:
>> http://www.freelists.org/archives/oracle-l/07-2006/msg00260.html
>>
>> I did not dare to set any hidden parameters yet; perhaps those would
>> have boosted performance even more.
>>
>> Can you confirm that reverting mentioned parameter to the non-default
>> value (FALSE) improves performance so dramatically?
>>
>> What is the rationale of such detrimental default value of the
>> initalization parameter that effectively renders the database
>> "misconfigured"?
>>
>> Are there any other deadly preset parameters I should be aware of
>> after Upgrade to Oracle 10g, according to your experience?
>>
>> Do you really recommend to let Oracle do things automatically? (SGA,
>> PGA, db_file_multibloc_read_count, etc.)
>
>I guess with this information given you cannot expect profound help.
>What did you do to pinpoint the source of the slowness? I mean, do you
>*know* where time is spent? If not, I'd investigate this first before
>doing any changes to the system.
>
>Kind regards
>
> robert