One interesting support document has to do with a bug related to adaptive log
file sync being enabled in 11.2.0.3. See document 1462942.1 hangs related to
adaptive log file sync - describes bug 13930580.
Also document 1541136.1 provides a bit more information about the adaptive log
file sync method, which also highlights a couple of additional bugs related to
this new 11gR2 feature.
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On ;
Behalf Of Mladen Gogala
Sent: Wednesday, May 18, 2016 8:35 PM
To: oracle-l@xxxxxxxxxxxxx
Subject: Re: LOG_BUFFER
On 05/17/2016 02:59 PM, Ram Cheruvattath wrote:
Hi
We have a 12c database that has very high redo generation (4TB/hr). Our
log_buffer is set to 1GB.
Is there a downside to setting log_bugger this big?
Thanks
Ram
Ram, we have talked about multi-threaded log writer in 12c few days ago. In the
meantime, I did some digging and found some new undocumented parameters:
_adaptive_scalable_log_writer_enable_worker_threshold Increase in redo
generation rate as a percentage
_adaptive_scalable_log_writer_disable_worker_threshold Percentage of overlap
across multiple outstanding writes
_adaptive_scalable_log_writer_sampling_count Evaluate single versus
scalable LGWR every N writes
_adaptive_scalable_log_writer_sampling_time Evaluate single versus
scalable LGWR every N seconds
_adaptive_scalable_log_writer_enable_worker_aging Per million of redo gen
rate when LGWR workers were last used
Of course, there is nothing about these parameters on the support.oracle.com.
The good, old times when it was possible to figure out how things work by
digging through the Oracle support documents are a long gone era. However, I
was able to dig out few things from Jonathan's site:
https://jonathanlewis.wordpress.com/2013/07/04/12c-trivia/
https://jonathanlewis.wordpress.com/2013/06/25/12c/
Not much more than we've already discussed, but still worth reading. I would
expect multi-threaded writer to be able to cope with much larger buffers than
we have seen in 11G. I don't think that in 12c having 1G log buffer is a bad
idea. However, the only sign you will get that your log buffer is too small are
log buffer space wait events in V$SYSTEM_EVENT. I guess that there is a feature
called "adaptive scalable log writer" which will appear in some kind of an
internals book when people at Oracle Corp. decide do tell someone about it.
--
Mladen Gogala
Oracle DBA
Tel: (347) 321-1217