Allocation of the 80-90% of resources for dbexport
Posted in 2007
A user on RHEL4/IDS 10 wanted to speed up a 2-hour 'dbexport -q' of a 40GB database, noting the CPU was 90% idle. Replies said dbexport is single-threaded and ASCII-based, so it can't be given more CPU; the bottleneck is likely disk/IO (check with iostat or sar -d), and writing to the journalled ext3 filesystem hurts. Suggested alternatives: use HPL, or parallel per-table UNLOADs with PDQ plus dbschema, or use ontape archives plus logical log backups for moving a database to another server. Side discussion clarified that on Linux 2.6 IDS opens block ('cooked') devices with O_DIRECT, making raw devices unnecessary. No benchmark result or final outcome was reported by the original poster.
Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.
Topics: Migration, Import/Export & Data Conversion
Hi, I'm using RHEL4/EMT64T and IDS10. There is a database with 40 GB size. The
'dbexport -q' takes about 2 hours. But the result of the sar is 90 %idle. I
would like to speed up that process anyway. How to allocate more resources for
dbexport? Or there are any methods/recomendations to reduce the backup time?
RUSTEM GISATULLIN said:
> Hi, I'm using RHEL4/EMT64T and IDS10. There is a database with 40 GB size.
> The
> 'dbexport -q' takes about 2 hours. But the result of the sar is 90 %idle.
> I
> would like to speed up that process anyway. How to allocate more resources
> for
> dbexport? Or there are any methods/recomendations to reduce the backup
> time?
It sounds like your problem is more on the disk side. Are you exporting to
a journalled filesystem, like ReiserFS?
--
Bye now,
Obnoxio
"I'm astonished anyone pays real money for this crap."
-- Cosmo
--
This message has been scanned for viruses and
dangerous content by OpenProtect(http://www.openprotect.com), and is
believed to be clean.
On 21/03/07, Obnoxio The Clown <obnoxio@serendipita.com> wrote:
> RUSTEM GISATULLIN said:
> > Hi, I'm using RHEL4/EMT64T and IDS10. There is a database with 40 GB size.
> > The
> > 'dbexport -q' takes about 2 hours. But the result of the sar is 90 %idle.
> > I
> > would like to speed up that process anyway. How to allocate more resources
> > for
> > dbexport? Or there are any methods/recomendations to reduce the backup
> > time?
>
> It sounds like your problem is more on the disk side. Are you exporting to
> a journalled filesystem, like ReiserFS?
>
> --
> Bye now,
> Obnoxio
>
> "I'm astonished anyone pays real money for this crap."
> -- Cosmo
>
> --
Why are you using dbexport as your backup mechanism? You can only
restore to this point in time (no roll forward of logged
transactions), and you are stopping users accessing your data while
the archive takes place. Also you are forcing the engine to pull the
data off in ASCII (text) format, which is slow, rather than page
images.
Keith
I'm use ext3 filesystem. I need this backup mechanism for restoring database
on other sides/servers.
Is it possible to allocate more resources for dbexport process namely?
You should use HPL to speed up your export or many unload together (one for
each table) and dbschema. using unload you can use PDQ.
dbexport is just a process, to use more cpu resources depends on how many cpu
do you have. if you have many cpu, the dbexport will use only one.
maybe you have a disk bottleneck.
you should use iostat or sar -d to check I/O.
Best Regards,
Celso Coimbra
-----Mensagem original-----
De: ids-bounces@iiug.org [mailto:ids-bounces@iiug.org]Em nome de RUSTEM
GISATULLIN
Enviada em: quarta-feira, 21 de março de 2007 07:51
Para: ids@iiug.org
Assunto: Re: Allocation of the 80-90% of resources for .... [8688]
I'm use ext3 filesystem. I need this backup mechanism for restoring database
on other sides/servers.
Is it possible to allocate more resources for dbexport process namely?
*******************************************************************************
Forum Note: Use "Reply" to post a response in the discussion forum.
Databases on journaling filesystems is a VERY bad idea. IDS and most RDBMSes
perform best on RAW devices (on Linux using O_DIRECT on COOKED devices seems to
perform equivalently and the latest releases of 9.40, 10.00, & 11.00 use
O_DIRECT on Linux, so COOKED devices with no filesystems are OK on Linux).
Use ontape archives and logical log backups to restore the server to another
machine. It's safer and more reliable anyway.
Art S. Kagel
----- Original Message -----
From: Rustem Gisatullin <ids@iiug.org>
At: 3/21 6:52:03
I'm use ext3 filesystem. I need this backup mechanism for restoring database
on other sides/servers.
Is it possible to allocate more resources for dbexport process namely?
*******************************************************************************
Forum Note: Use "Reply" to post a response in the discussion forum.
I mean, I'm using raw devices of course for all database spaces. The
file system of ext3 type is just location for unloaded files by
dbexport.
Sorry for my question, but tell me please, how and where to set up O_DIRECT
flag for cooked devices.
IDS does this (opens the chunk files with the O_DIRECT flag set) automatically
in the latest releases. You don't have to do anything. I don't remember what
version of IDS you're running, but IBM tech support can certainly tell you if
your release has that feature or if you should upgrade to obtain it.
Art S. Kagel
----- Original Message -----
From: Rustem Gisatullin <ids@iiug.org>
At: 3/21 10:40:26
I mean, I'm using raw devices of course for all database spaces. The
file system of ext3 type is just location for unloaded files by
dbexport.
Sorry for my question, but tell me please, how and where to set up O_DIRECT
flag for cooked devices.
*******************************************************************************
Forum Note: Use "Reply" to post a response in the discussion forum.
Thanks a lot, I've got IDS 10.00.FC3R1. But I need to perform any benchmark before use cooked files rather than raw devices.
It would be prudent to test this yourself, yes. Art S. Kagel ----- Original Message ----- From: Rustem Gisatullin <ids@iiug.org> At: 3/21 11:22:00 Thanks a lot, I've got IDS 10.00.FC3R1. But I need to perform any benchmark before use cooked files rather than raw devices. ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.
As a note, ext3 is a journaling file system. You could still be seeing
a performance hit due to that.
On Wed, 2007-03-21 at 05:51 -0500, RUSTEM GISATULLIN wrote:
> I'm use ext3 filesystem. I need this backup mechanism for restoring database
> on other sides/servers.
>
> Is it possible to allocate more resources for dbexport process namely?
>
>
>
*******************************************************************************
> Forum Note: Use "Reply" to post a response in the discussion forum.
>
--
--------------------------------------------
Chris Salch
Programmer/Analyst
LeTourneau University
903-233-3537
You wrote earlier "..on Linux using O_DIRECT on COOKED devices seems to perform equivalently and the latest releases of 9.40, 10.00, & 11.00 use O_DIRECT on Linux, so COOKED devices with no filesystems are OK on Linux..." Please explain what do you mean - "so COOKED devices with no filesystems". How to create cooked files without filesystems?
=46or Linux kernels 2.6 or later the use of 'raw' command is outdated. But= =20 most Linux distributions still maintain it for compatibility. IDS opens=20 the block devices with O_DIRECT flag and the kernel use it like raw=20 devices. O_DIRECT flag was introduced in IDS 10.0 with KAIO. See more in=20 Sandor's developerWorks article at=20 http://www-128.ibm.com/developerworks/db2/library/techarticle/dm-0503szabo/= #N100D0 Andreas On Thursday 22 March 2007 06:54, RUSTEM GISATULLIN wrote: > You wrote earlier "..on Linux using O_DIRECT on COOKED devices seems > to perform equivalently and the latest releases of 9.40, 10.00, & > 11.00 use O_DIRECT on Linux, so COOKED devices with no filesystems > are OK on Linux..." > > Please explain what do you mean - "so COOKED devices with no > filesystems". How to create cooked files without filesystems? > > > ********************************************************************* >********** Forum Note: Use "Reply" to post a response in the > discussion forum. =2D-=20 Andreas Breitfeld; Informix Development Munich IBM Deutschland GmbH; Vorsitzender des Aufsichtsrats: Hans Ulrich=20 Maerki; Gesch=C3=A4ftsf=C3=BChrung: Martin Jetter (Vorsitzender), Rudolf Ba= uer,=20 Christian Diedrich, Christoph Grandpierre, Matthias Hartmann, Andreas=20 Kerstan; Sitz der Gesellschaft: Stuttgart; Registergericht: Amtsgericht=20 Stuttgart, HRB 14562; WEEE-Reg.-Nr. DE 99369940
hopefully following text looks better: =46or Linux kernels 2.6 or later the use of 'raw' command is outdated. But most Linux distributions still maintain it for compatibility. IDS opens the block devices with O_DIRECT flag and the kernel use it like raw devices. O_DIRECT flag was introduced in IDS 10.0 with KAIO. See more in Sandor's developerWorks article at http://www-128.ibm.com/developerworks/db2/library/techarticle/dm-0503szabo/= #N100D0 Andreas On Thursday 22 March 2007 06:54, RUSTEM GISATULLIN wrote: > You wrote earlier "..on Linux using O_DIRECT on COOKED devices seems > to perform equivalently and the latest releases of 9.40, 10.00, & > 11.00 use O_DIRECT on Linux, so COOKED devices with no filesystems > are OK on Linux..." > > Please explain what do you mean - "so COOKED devices with no > filesystems". How to create cooked files without filesystems? > > > ********************************************************************* >********** Forum Note: Use "Reply" to post a response in the > discussion forum. =2D-=20 Andreas Breitfeld; Informix Development Munich IBM Deutschland GmbH; Vorsitzender des Aufsichtsrats: Hans Ulrich=20 Maerki; Gesch=C3=A4ftsf=C3=BChrung: Martin Jetter (Vorsitzender), Rudolf Ba= uer,=20 Christian Diedrich, Christoph Grandpierre, Matthias Hartmann, Andreas=20 Kerstan; Sitz der Gesellschaft: Stuttgart; Registergericht: Amtsgericht=20 Stuttgart, HRB 14562; WEEE-Reg.-Nr. DE 99369940
COOKED or block device files - as opposed to filesystem files or RAW (character) devices. Art S. Kagel ----- Original Message ----- From: Rustem Gisatullin <ids@iiug.org> At: 3/22 1:54:46 You wrote earlier "..on Linux using O_DIRECT on COOKED devices seems to perform equivalently and the latest releases of 9.40, 10.00, & 11.00 use O_DIRECT on Linux, so COOKED devices with no filesystems are OK on Linux..." Please explain what do you mean - "so COOKED devices with no filesystems". How to create cooked files without filesystems? ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.