Re: Re: AW: Re: Large index creation speed up / issues with direct path read temp / direct path write temp

From: Stojan Veselinovski <stojan.veselinovski_at_gmail.com>
Date: Mon, 06 Jul 2015 00:06:52 +0000
Message-ID: <CALn1tDvuKbGpCn-cn6OGsqQAE2GWCX0iraGug9KnVnztZ9-Bvw_at_mail.gmail.com>

A few years ago I had a performance issue where I was seeing high latency on direct path reads and writes to temp on a T4 LDOM. I tuned the query to get around the issue but the general feel at the time was that the T4's weren't performing as well as our older Intel boxes.

I put together a small test case and compared environments and found that direct path reads or writes were approx 20 (? going of memory) times slower on the T4 than the physical Intel boxes.

All were hitting the same SAN.

I never got to the bottom of it as I left the company but this mail reminded me of the issues I experienced. I'll see if I can dig up some doco on it.

Regards,

Stojan
www.stojanveselinovski.com/blog

On Mon, Jul 6, 2015 at 1:24 AM l.flatz <l.flatz_at_bluewin.ch> wrote:

> Hi Stefan
>
> you are right. It scares me too.
> Guess there will be a way, can't be the first time something like that
> happens.
>
>
> Von meinem Samsung Gerät gesendet.
>
>
> -------- Ursprüngliche Nachricht --------
> Von: Stefan Koehler <contact_at_soocs.de>
> Datum: 05.07.2015 12:19 (GMT+01:00)
> An: oracle-l_at_freelists.org, GG <grzegorzof_at_interia.pl>, l.flatz_at_bluewin.ch
> Betreff: Re: AW: Re: Large index creation speed up / issues with direct
> path read temp / direct path write temp
>
> Hi GG,
> i am also a performance specialist, but had to deal with such situations
> in my "permanent employment" era.
>
> _at_Lothar:
> Unfortuanately it is not so easy in case of physical standby databases :)
> In addition these databases usually run in "force logging" mode so that a
> "create index .... nologging" would have no effect anyway. There are some
> special cases (like BI ETLs, etc.) where nologging is also an option in
> case of physical standby database, but this is not the common stuff.
>
> However if you just run this index creation with nologging (and database
> is not in force logging mode), you can use RMAN incremental backups to roll
> forward a physical standby database:
> http://docs.oracle.com/cd/E11882_01/server.112/e41134/rman.htm#SBYDB00759
>
> Almost anything is possible with Oracle - scary :)
>
> Best Regards
> Stefan Koehler
>
> Freelance Oracle performance consultant and researcher
> Homepage: http://www.soocs.de
> Twitter: _at_OracleSK
>
> > "l.flatz" <l.flatz_at_bluewin.ch> hat am 5. Juli 2015 um 11:45 geschrieben:
> >
> > Oh, I see. Thanks for letting me know. I am a performance specialist.
> Therefore I am not aware how to deal with this situation. Maybe our
> > colleagues will give you some advice.
> > I would simply create the index on both sides which is probably pretty
> naive. ☺
> >
> > -------- Ursprüngliche Nachricht --------
> > Von: GG <grzegorzof_at_interia.pl>
>
> > Datum: 05.07.2015 11:07 (GMT+01:00)
> > An: l.flatz_at_bluewin.ch, oracle-l_at_freelists.org
> > Betreff: Re: Large index creation speed up / issues with direct path
> read temp / direct path write temp
> >
> > > > Standby involved (production HA) no way :( .
>
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>

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

Received on Mon Jul 06 2015 - 02:06:52 CEST