Results 1 to 5 of 5
  1. #1
    Join Date
    Mar 2005
    Posts
    73

    Unanswered: DirectI/0 Vs Devices.

    All-
    I would like to know the advantage of having File containers with Direct I/o over a device containers.

    some of our systems built by IBM and they proeferred a file containers over devices and they claim that Direct I/o feature made file containers better than device containers.

    Is any one having experience with Direct I/O File containers and is it really better than Device containers.

    please share your ideas.

    thanks

  2. #2
    Join Date
    Sep 2003
    Posts
    237
    File containers are easy to create and manipulate(copy, expand etc.). With file containers I/O is done from disk to OS file cache and then from OS file cache to bufferpool. With direct I/O (as with device)data goes from disk to bufferpool. For the same memory on a system you can use bigger bufferpools and smaller OS file cache in Direct I/O or device(raw) containers making the sytem more efficient. With Direct I/O, You get the convenience of the file system with the efficiency of device container. You have to adjust the OS parameters.
    mota

  3. #3
    Join Date
    Mar 2005
    Posts
    73
    Thank you DBA MOTA. I know the differences . But what do you think is the best between these , Before DIO , CIO -- devices are easily better than Files.

    what about now. Does devices still performing better than File containers with DIO and CIO..

    thanks

  4. #4
    Join Date
    May 2003
    Location
    USA
    Posts
    5,737
    Unless most of SQL statements are doing table scans on large tables, it will not make any difference.
    M. A. Feldman
    IBM Certified DBA on DB2 for Linux, UNIX, and Windows
    IBM Certified DBA on DB2 for z/OS and OS/390

  5. #5
    Join Date
    Jan 2007
    Location
    Jena, Germany
    Posts
    2,721
    If you have BLOB data (on LUW), you want to use file-based containers and activate file system caching for those. For everything you don't want to have the OS apply an additional caching layer. That's all I would worry about because it will be really hard for you to measure any differences between the types - as Marcus_A said. (And if you have lots of table scans, you have other mechanisms to improve performance first.)
    Knut Stolze
    IBM DB2 Analytics Accelerator
    IBM Germany Research & Development

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •