Re: db_block_checkpoint

From: Steve Adams <steveadams_at_acslink.net.au>
Date: 1998/03/09
Message-ID: <35046ec0.2852740736@newsserver.trl.oz.au>#1/1

Hello Brian,

The default value for db_block_checkpoint_batch is 8 and the maximum is db_files * db_file_simultaneous_writes / 2 up to a maximum of db_block_buffers / db_block_lru_latches / 4 or a platform specific limit (commonly 512), whichever is the lesser. The impact of setting db_block_checkpoint_batch to its maximum value is that DBWR will focus ALL its effort on completing the checkpoint as quickly as possible to the detriment of its normal cleaning of blocks from the write list. This means that no free blocks will be returned to the LRU end of the LRU chains for the duration of the checkpoint, increasing the risk and duration of 'free buffer waits' significantly. Therefore it is inadvisable to just set this parameter to a high value. A reasonable starting point is to set start at 1/4 of the maximum value for your platform, and increase incrementally if the 'Checkpoint not complete' messages are not eliminated, or decrease by factors of 2 if 'free buffer waits' are introduced. In most case, this means that you should start with db_block_checkpoint_batch set to 2 * db_files.

Hope this helps,
Regards,
Steve Adams
steveadams_at_acslink.net.au

On Fri, 06 Mar 1998 08:48:54 -0600, bdg_at_hotmail.com wrote:

>I'm trying to eliminate "Checkpoint not complete." messages in
>the alert log. The alerts are most prevalent during a SQL*Loader
>run. I'm up to 1.4 GB of online redo logs (I really want to
>lower this to about .5 GB) and have the checkpoint process running.
>I came across the db_block_checkpoint_batch init.ora parameter in
>the Oracle Performance Tuning book (O'Reily & Assoc). It seems to
>be relevant to the problem, but there was no indication of what
>would be considered a relatively high value. Has anyone
>successfully applied this parameter and if so, what was found to
>be an appropriate value?
>
>Any sort of heuristic such as doubling the value, halves the
>amount of time taken to to complete a checkpoint would also be
>helpful.
Received on Mon Mar 09 1998 - 00:00:00 CST

Re: db_block_checkpoint_batch