Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.server -> Re: ORACLE on Linux - IO bottleneck
Eder San Millan wrote:
>
> I can suppose you are saying that in 2.4 kernels, RAWDEVICES only
> allows 512 byte IOs ??
>
low level? Yep.
>> The first case show direct I/O (syncronous). >> The second one a buffered I/O.
No. I used a dd and the worload was syncronous. Oracle can exploit the kaio capabilities of linux (search for libaio). The difference is in the sycalls used.
>
> So, could we say that RawDevices in 2.4 kernels are not recomended or
> this is supposing too much. We will upgrade to 2.6 kernel but this
> suppose us many problems that must solve with a lot of time and this
> other problem is being done in a productive system and....well...you
> know ...
>
****Test on filesystem before going with 2.6.**** Check if there are performance differences. I used rawdevices on top of LVM on a RAC 9i for a couple of years with good performance (and it was an oooold SLES7).
>> What do you mean saying that your device had 1k block size?!? >> Isn't you using it "raw"?
Personally I don't know the --getbsz. I cannot find it in the man pages
on the web.
Tomorrow, from office, I'm goign to check the sources of blockdev.
I see there is even setbsz. Have yuo tried changing it?
> ----------------------------------------------------------------
> SDC is the device that Storage system present me, this "disk" is part
> of a VG that we divide in many LVs where we link the rawdevices
> This value (bsz=1024) isn't important?
>
Yes, it makes a lot of different (sorry, I didn't ask you if you were
using LVM).
On LVM1, lvcreate can specify the readahead value and the "contiguity"
(plus tons of other parameters).
I believe now it is time for benchmarking. :)
Maybe tomorrow I can give you more information.
-- Fabrizio Magni fabrizio.magni_at_mycontinent.com replace mycontinent with europeReceived on Thu Feb 09 2006 - 15:57:58 CST