RE: On AIX using DIO/CIO on JFS2
- From: Denny Koovakattu <oracle-l@xxxxxxxxxxxxxx>
- To: Brandon.Allen@xxxxxxxxxxx
- Date: Tue, 21 Feb 2006 15:43:10 -0600
Hi Brandon,
I agree that under normal circumstances, this should not be an issue for
Oracle datafiles. But just because the filesystem is configured to use DIO/CIO
does not mean all IO against that filesystem will use DIO/CIO . Unless certain
conditions (alignment, open mode) are met, it will revert to normal cached IO.
Also if any process has a file open in an incompatible mode (non DIO/CIO mode),
IO against this file from other processes will revert to cached IO until the
process which has the file open in an incompatible mode closes the file. Unless
we can guarantee that this will not happen, it would be prudent to patch the
system. For instance, what would be the effect of say using cp against a
datafile for taking an old fashioned backup ? We did not have any issues with
this bug so far.
Thanks,
Denny
--
Denny Koovakattu
Quoting "Allen, Brandon" <Brandon.Allen@xxxxxxxxxxx>:
> Thanks for the info, but I'm not sure how this would cause problems for
> Oracle. If you're using DIO or CIO, how would you get a "cached write" since
> dio/cio bypass the cache by definition? Also, if you're working with Oracle
> blocks of say 8k, and your cio filesystem is created with the default 4k page
> size, how would you get a non-page-aligned file size?
>
> Did you actually have problems in Oracle due to this bug?
>
> Thanks,
> Brandon
----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.
--
//www.freelists.org/webpage/oracle-l
Other related posts:
