Hi, Remigiusz,
Some more ideas to consider:
- if you suspect bad performance of the storage containing your redo
logs (case B in Cary Millsap's notation), it would be helpful to check
'log file parallel writes' wait event. This event is recorded and IO
time is measured by LGWR itself, so this is reliable source of info at
first hand. The only problem with this measure is that the IO size may
vary significantly... Though in your case (frequent commits) that would
not be the case.
- the problem might also be with OS delays in redo log file opening at
log switch. Why not to try to help OS and make this process a little bit
more easy? Steve Adams has a script on his site named hold_logs_open.sh.
IMHO, it's worth looking at.
- well, you can also turn on LGWR tracing at level 8 to see log switch
details in the microscope...
HTH
--Andrey
Remigiusz Soko?owski wrote:
> Thanks for all suggestions and explaination of the whole process
--
http://www.freelists.org/webpage/oracle-l
Received on Sat Feb 24 2007 - 09:34:18 CST
