Results 1 to 3 of 3
  1. #1
    Join Date
    Sep 2003
    Location
    Burlington, MA
    Posts
    1

    Unanswered: experienceing HIGH disk utilization from redo

    Hi All,

    I have an application running on Weblogic that requires session persitence. I am running Oracle 9.2.0. The table created to store these sessions (which are quite large - 500K) looks like this:

    ( wl_id VARCHAR2(100) NOT NULL,
    wl_context_path VARCHAR2(100) NOT NULL,
    wl_is_new CHAR(1),
    wl_create_time NUMBER(20),
    wl_is_valid CHAR(1),
    wl_session_values LONG RAW,
    wl_access_time NUMBER(20),
    wl_max_inactive_interval INTEGER,
    PRIMARY KEY (wl_id, wl_context_path) )

    I created a seperate database for this table. When applying load onto my application, the symptoms I see are:

    high buffer busy waits (data block and segment header)
    over 98% disk utilization on the physical disk that the redo logs reside on
    log buffer space waits

    Note that this table get a high volume of update and inserts. However, this data does not ever need to be archived or rolledback. In fact, Weblogic will delete these rows after 15 minutes.

    My question is, how can I tune this? What things should I look for? I have already tried adjusting:

    Freelists, log_buffer, db_cachesize, db_block_size

    Would it be possible remove all logging (redo, undo) for this table? Would creating it in a temporary tablespace do the trick?

    Any advince would be greatly appreciated. Thanks!

    -Jack

  2. #2
    Join Date
    Mar 2002
    Location
    Reading, UK
    Posts
    1,137
    OK you need to reduce the number of rows per block to reduce contention AND if you have any indexes then you have to make sure there isnt a hotspot on the end of the index due to wl_id (for example if wl_id is a number and increments by one then all the inserts would hit the same part of the tree and associated leaf nodes).

    So 1) reduce pctused and increase pctfree to cut the number of rows stored in a block. Also increase freelists & initrans.

    2) If wl_id does cause hotspots on indexes then specify reverse on the index to reverse the byte order.

    Alan

  3. #3
    Join Date
    Mar 2002
    Location
    Reading, UK
    Posts
    1,137
    For tuning redo logs try these links

    http://www.ixora.com.au/notes/
    http://www.ixora.com.au/tips

    Alan

Posting Permissions

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