Re: log writer tuning
- From: "Anand Rao" <panandrao@xxxxxxxxx>
- To: rems@xxxxxxxx
- Date: Thu, 22 Feb 2007 10:46:51 +0530
Have you thought about raw volumes? i didn't see any suggestions for that,
so if i am repeating, my apologies.
redologs on raw is really what you should have. Not sure what RAID levels
you have at your disposal, but if you can do RAID 1+0 or 0+1 (don't want
start another RAID war here!). if you don't have that luxury, RAID 5 is
acceptable as long as you have more disks and plan it correctly. of course,
you should move the redologs away from the datafiles. move away from
filesystems!
if you are using ASM, i guess there is bit more involved there.
What you really need is more spindles to do your writes. how you stripe
them, blah blah is another huge discussion. try raw, will help.
talking of writes, what is your redo size? you can look at that in your AWR
or Statspack. maybe some jazzy GUI tool will also help.
find the size of redo per second. that will give you a figure (average)
about the size of your writes. Then, (re)size your log buffer in that range
and then start measuring again. what you should watch out for is redo bursts
and sustained redo rates.
a lot depends on the rate of redo generation. so, if you have some good idea
about the application, how often it commits, etc. that will *really* help.
regards
anand
On 22/02/07, Mladen Gogala <mgogala@xxxxxxxxxxx> wrote:
On 02/21/2007 04:29:32 PM, Kevin Closson wrote:
> few Oracle ports where LGWR can post multiple waiters with one systems
> call (certainly not Larry's Linux). Remember, every time a process comes
> out of kernel mode, the opportunity arises to schedule a higher priority
Speaking of Linux, is anybody out there using Adaptec RAID with I2O
drivers?
I20 drivers should, in theory, drastically increase the throughput and
significantly
reduce CPU usage by cutting down on the number of interrupts.
--
Mladen Gogala
http://www.mladen-gogala.com
--
//www.freelists.org/webpage/oracle-l
Other related posts:
