4 Replies Latest reply: Apr 8, 2013 9:55 AM by 1001350 RSS

    FAST_START_MTTR_TARGET parameter advice


      I need some advice in setting the FAST_START_MTTR_TARGET initialization parameter. There is a lot of valuable documentation and blog posts out there but I am unable to find concrete guidance on this.

      In reading the Checkpoint Tuning and Troubleshooting Guide (147468.1), there is mention of 4 total parameters (including FAST_START_MTTR_TARGET) that need to be visited in order to set this:


      My current db server environment settings are as follows:

      W2K8-R2-Std, 11g-
      OS-block-size = 4096 bytes

      The current initialization parameter settings are:


      The current redo log files are set at 51,200KB.

      The condition warranting this is these entries in the alert log:

      Thread 1 cannot allocate new log, sequence 117424
      Private strand flush not complete
      Current log# 1 seq# 117423 mem# 0: K:\ORACLE_SID\REDOG1M1.LOG
      Current log# 1 seq# 117423 mem# 1: L:\ORACLE_SID\REDOG1M2.LOG
      Thread 1 advanced to log sequence 117424 (LGWR switch)
      Current log# 3 seq# 117424 mem# 0: D:\APP\ORADATA\ORACLE_SID\REDOG3M1.LOG
      Current log# 3 seq# 117424 mem# 1: L:\ORACLE_SID\REDOG3M2.LOG

      Kindly advise. Thank you in advance.
        • 1. Re: FAST_START_MTTR_TARGET parameter advice
          The parameter required for checkpointing since 9i is FSMMT and from 10.2 onwards, if it is set to a non-zero value, it would be automatically tuned by Oracle. As long as you are not hitting the checkpoint not complete wait event, you are most probably okay. Also, if you would set the FSMMT , Oracle offers a redo log file size advisor which can help in finding the right size of the redo log for your db.

          Just my 2 cents.

          • 2. Re: FAST_START_MTTR_TARGET parameter advice
            Thank you Aman.
            What would you recommend be the starting threshold for the FSMMT parameter? Would this also require an increase to the db_writer_processes parameter?

            • 3. Re: FAST_START_MTTR_TARGET parameter advice
              Look in your alert log to see how long it is between redo switches during times of heavy usage. Multiply the size of the redo by how much it would take to switch every 30 minutes. For example, if you switch every 5 minutes, multiply by 6. Then make them much larger and rely on the 1800 second timeout (or better, do the figuring during maximal loading or batch update times). Let it run a few days then check the advisor as in http://docs.oracle.com/cd/E25178_01/server.1111/e16638/build_db.htm#autoId3

              Here's the proper way to figure FSMT http://docs.oracle.com/cd/E25178_01/server.1111/e16638/instance_tune.htm#PFGRF13015

              Don't bother changing dbw settings yet. Wait until you see some evidence you need to do that.
              • 4. Re: FAST_START_MTTR_TARGET parameter advice
                Thank you jgarry.