FAQ
I found out on a database (8174, Solaris 8 Sparc, EMC) that I use,
while disk_asynch_io is set to true, the dbwr_io_slaves is set to non zero, 6.
And I am seeing a lot of "slave wait" events. The two obviously conflicts and
I wonder what kind of effect this have caused when both are enabled.

Thanks.

Search Discussions

  • Jared Still at Dec 8, 2004 at 10:59 am
    I haven't worked on Solaris in awhile, so I don't n know
    if the following is true on your version of Solaris or not. I
    believe I last tested it on 2.6 or 2.7.

    Enabling disk_io_sync on Solaris does not not cause
    async IO on filesytems, only on raw disk.

    You may want to investigate.

    Jared
    On Mon, 6 Dec 2004 22:21:10 -0500, Richard Ji wrote:
    I found out on a database (8174, Solaris 8 Sparc, EMC) that I use,
    while disk_asynch_io is set to true, the dbwr_io_slaves is set to non zero, 6.
    And I am seeing a lot of "slave wait" events. The two obviously conflicts and
    I wonder what kind of effect this have caused when both are enabled.

    Thanks.
    --
    http://www.freelists.org/webpage/oracle-l
    --
    Jared Still
    Certifiable Oracle DBA and Part Time Perl Evangelist
    --
    http://www.freelists.org/webpage/oracle-l
  • Richard Ji at Dec 8, 2004 at 11:26 am
    Jared,

    As far as I understand it. On Solaris 8 and onward, by default
    Asynch IO is turned on and Oracle will use it by issue a KAIO. If it's
    raw device or Veritas quick IO, the KAIO will succeed. Otherwise
    KAIO will fail and an AIO that's simulated by LWP doing pwrite() will
    happen. That's why there are 258 LWP under the DBWR process.

    Now, I can turn Asynch IO off and use Oracle's dbwr slaves to
    simulate that as well. But as far as I know, one would only want
    to do that on a platform that doesn't support AIO. In either case,
    you chose one way or the other. But on this DB that I saw, both
    are set. So I am wondering which one will it use? I don't have
    enough permission on the system to truss the process or anything.
    So I wonder if anyone has seen it before and the implication of
    having both set.

    Best regards

    Richard Ji
    On Wed, 8 Dec 2004 08:58:49 -0800, Jared Still wrote:
    I haven't worked on Solaris in awhile, so I don't n know
    if the following is true on your version of Solaris or not. I
    believe I last tested it on 2.6 or 2.7.

    Enabling disk_io_sync on Solaris does not not cause
    async IO on filesytems, only on raw disk.

    You may want to investigate.

    Jared



    On Mon, 6 Dec 2004 22:21:10 -0500, Richard Ji wrote:
    I found out on a database (8174, Solaris 8 Sparc, EMC) that I use,
    while disk_asynch_io is set to true, the dbwr_io_slaves is set to non zero, 6.
    And I am seeing a lot of "slave wait" events. The two obviously conflicts and
    I wonder what kind of effect this have caused when both are enabled.

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

    --
    Jared Still
    Certifiable Oracle DBA and Part Time Perl Evangelist
    --
    http://www.freelists.org/webpage/oracle-l
  • Jared Still at Dec 8, 2004 at 12:45 pm
    Ah, good question.

    I have access to Solaris, but not the correct version.
    On Wed, 8 Dec 2004 12:26:21 -0500, Richard Ji wrote:
    Jared,

    As far as I understand it. On Solaris 8 and onward, by default
    Asynch IO is turned on and Oracle will use it by issue a KAIO. If it's
    raw device or Veritas quick IO, the KAIO will succeed. Otherwise
    KAIO will fail and an AIO that's simulated by LWP doing pwrite() will
    happen. That's why there are 258 LWP under the DBWR process.

    Now, I can turn Asynch IO off and use Oracle's dbwr slaves to
    simulate that as well. But as far as I know, one would only want
    to do that on a platform that doesn't support AIO. In either case,
    you chose one way or the other. But on this DB that I saw, both
    are set. So I am wondering which one will it use? I don't have
    enough permission on the system to truss the process or anything.
    So I wonder if anyone has seen it before and the implication of
    having both set.

    Best regards

    Richard Ji
    --
    Jared Still
    Certifiable Oracle DBA and Part Time Perl Evangelist
    --
    http://www.freelists.org/webpage/oracle-l
  • Tim Gorman at Dec 23, 2004 at 10:44 pm
    Richard,

    DBWR_IO_SLAVES is a really poor attempt at asynch I/O, and it has dozens of
    bugs against it from v8.0.x through v9.2.0x. Don't take my word for it --
    search MetaLink on the parameter name as keyword, be sure to click on
    "Advanced Search" and "Bug Database". It is a bottleneck to performance
    besides. Don't use it.

    If you feel you need more bandwidth on your DBWR, then please consider
    setting DB_WRITER_PROCESSES > 1 -- setting it no higher than CPU_COUNT seems
    to work OK.

    Hope this helps...

    -Tim
    on 12/6/04 8:21 PM, Richard Ji at richard.c.ji_at_gmail.com wrote:

    I found out on a database (8174, Solaris 8 Sparc, EMC) that I use,
    while disk_asynch_io is set to true, the dbwr_io_slaves is set to non zero, 6.
    And I am seeing a lot of "slave wait" events. The two obviously conflicts and
    I wonder what kind of effect this have caused when both are enabled.

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

Related Discussions

Discussion Navigation
viewthread | post
Discussion Overview
grouporacle-l @
categoriesoracle
postedDec 6, '04 at 9:25p
activeDec 23, '04 at 10:44p
posts5
users3
websiteoracle.com

People

Translate

site design / logo © 2022 Grokbase