AIX

 View Only
  • 1.  Oracle redo on SSD

    Posted Thu September 24, 2015 05:31 PM

    Originally posted by: BenKim


    Team,

     

    I'd like to seek your expertise on this one..

    I have an S824 server with Oracle 12.1.0.2 hoping to take a big performance advantage on "commits" by putting redologs on SSDs.

    I have been using Swingbench OE to generate load and redo performance has not been meeting my expectation and I wonder why.

    SQL> show parameter filesystemio

     

    NAME                                 TYPE        VALUE

    ------------------------------------ ----------- ------------------------------

    filesystemio_options                 string      setall

    oracle[/home/oracle] > mount

      node       mounted        mounted over    vfs       date        options      

    -------- ---------------  ---------------  ------ ------------ --------------- 

             /dev/fslvxx      /u04             jfs2   Sep 23 14:29 rw,cio,log=/dev/loglv00

             /dev/fslvxx      /u03             jfs2   Sep 23 14:29 rw,cio,log=/dev/loglv01

             /dev/fslvxx      /u06             jfs2   Sep 23 14:29 rw,cio,log=/dev/loglv03

             /dev/fslvxx      /u05             jfs2   Sep 24 11:35 rw,cio,log=/dev/loglv02

     

    after running OrderEntry for about 10 minutes, I see "log file sync" as a top wait event with above 5 ms of service time.

       +62  Top 10 Foreground Events by Total Wait Time

       +63  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

       +64                                             Total Wait       Wait   % DB Wait

       +65  Event                                Waits Time (sec)    Avg(ms)   time Class

       +66  ------------------------------ ----------- ---------- ---------- ------ --------

       +67  log file sync                    3,011,985        17K       5.63   18.6 Commit

       +68  DB CPU                                          12.9K              14.2

       +69  direct path write temp           1,886,757      10.6K       5.61   11.6 User I/O

       +70  db file sequential read          1,934,538     7571.9       3.91    8.3 User I/O

       +71  library cache: mutex X           3,405,352     2201.1       0.65    2.4 Concurre

       +72  log file switch (checkpoint in       2,178     1030.4     473.08    1.1 Configur

     

    So..any folks have used redo on SSDs?  Did it meet your expectations?

    Thank you very much for your help!

     

    Ben



  • 2.  Re: Oracle redo on SSD

    Posted Mon October 26, 2015 07:49 PM

    Originally posted by: selvamurugan


    What is the jfs2 block size of these redo filesystem, if it is set to default (4096) re-built them with 512. 



  • 3.  RE: Oracle redo on SSD

    Posted 20 days ago

    Ben,

    Firs, it is normally recommended have your redo logs on a different filesystem than your normal data. I know by default Oracle will create the redo logs on the same filesystem as your DB, but you would have to manually create different filesystems for your redo logs.

    Second, as @selvamurugan recommended, the JFS2 block size for the redo log filesystems should be set to 512 (instead of the default of 4096).

    The above should improve performance. You can also try to move the to SSDs, but I would first try the above changes, before you move to using SSDs.



    ------------------------------
    Abraham Alvarez
    ------------------------------