Frequency of Checkpoint

Question

Hi,
in 10g R2, where else other than Alertlog can I check the frequency of Checkpoint ?
In alertlog I have :
Thread 1 cannot allocate new log, sequence 3064
Checkpoint not complete
I can not see why. Any idea ?
Thank you.
I tried this :
ALTER SYSTEM SET fast_start_mttr_target=20 SCOPE=BOTH;
But in alertlog I saw :
Sun Nov 15 09:55:54 2009
ALTER SYSTEM SET fast_start_mttr_target=20 SCOPE=BOTH;
Sun Nov 15 09:56:10 2009
FAST_START_MTTR_TARGET 20 is out of the valid MTTR range, use 461 instead.
Sun Nov 15 09:57:31 2009

Answer 1

Hi,
The checkpoint frequency is affected by different configuration parameters. The most usual case is the size of redo logs. To use FSMT parameter (which is a straightforward way to control instance time to recover from instance crash and spikes of IO writes during checkpointing), you should use big enough redo logs - to reduce normal checkpoint frequency, and set FSMT to a reasonable value. For instance, suppose that you have 300 MB redo logs and your checkpoint frequency is once in 5 minutes due to log switch. To achieve frequency of once in 10 minutes, you should increase the size of redo logs to 600 MB. Then, if you'd like to make sure that DB crash recovery time is, let's say, 6 minutes, you should set FSMT to 360. Be aware that normal checkpoints due to log swich are not eliminated, they just becomes quicker due to less work to be done. Also remember that FSMT is some sort of "permanent checkpointing" - DBWR performs small writes to advance thread checkpoint, and does it all the time, thus reducing IO spikes by distributing number IO over the period.