Re: log buffer question

The log buffer is written out up to and including the most recent commit marker.

It is not surprising that log file syncs and log file parallel writes are similar, as a log file sync requires a log file write.

How many minutes per day are you losing
to log file writes, and what is the average wait time.

The counts may be quite high, but if the average time loss is very small, then the _apparent_ impact as far as the end-user is concerned may be irrelevant.

If you are getting hundreds of log switches per hour, you should probably look at
increasing the log file size (see the FAQ: http://www.jlcomp.demon.co.uk/faq/ind_faq.html )

and the log_checkpoint_interval in the init.ora.

It is possible that a high rate of switching could be forcing your database writer to be too busy, with the side-effect that file I/O is under pressure and slower than it ought to be - thus exaggerating the impact of the log file writes.

--
Jonathan Lewis

Seminars on getting the best out of Oracle
Last few places available for Sept 10th/11th
See http://www.jlcomp.demon.co.uk/seminar.html




Zin Zean wrote in message <3AE07A9C.87641F1A_at_yahoo.com>...


>Hello.

>I'm running 8.1.7 on Solaris 2.7.  My database supports a web site.

>

>When oracle writes to the redo log file, such as when a user performs a

>commit,

>does it write LOG_BUFFER (setting in the init.ora) number of bytes? ie.

>flush the entire log buffer?

>

>Or does it only write transaction's redo buffer information?

>

>The reason I'm asking is that the top two wait events in our system is

>log file sync and log file parallel writes.  Since these two events have

>waits

>on the same order of magnitude I'm assuming that we're writing to the

>redo log at a high rate b/c of the number of commits in our system.

>And we have LOG_BUFFER set to 512K and our redos at 10M we are

>seeing 100's of log switches in an hour.  And I'm hoping that if we

>reduce LOG_BUFFER the log switch frequency will go down.

>

>Thanks.

>

>

>