Ext3 users - Apr 2006 - ext3 data=ordered - good enough for oracle?

Given that the default journaling mode of ext3 (i.e. ordered), does not 
guarantee write ordering after a crash, is this journaling mode safe 
enough to use for a database such as Oracle?  If so, how are out of sync 
writes delt with?

Kind regards,

Herta

Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm

> Given that the default journaling mode of ext3 (i.e. ordered), does not 
> guarantee write ordering after a crash, is this journaling mode safe 
> enough to use for a database such as Oracle?  If so, how are out of sync 
> writes delt with?
Oracle manages its own I/O cache in userspace and handles data coherency related
to that. So data=journal is useless in this case.
I guess databases such as Oracle uses O_SYNC to control the flushing of data 
or even O_DIRECT to bypass the kernel cache.

Johann

Herta Van den Eynde wrote:
> Given that the default journaling mode of ext3 (i.e. ordered), does not
> guarantee write ordering after a crash, is this journaling mode safe
> enough to use for a database such as Oracle?
Yes, the database doesn't rely the kind of functionality that
data=journaled provides that data=ordered doesn't.  data=ordered is fine.
> If so, how are out of sync writes delt with?
The database, just like ext3/jbd, implements its own consistency
mechanisms by careful write ordering.  ext3 uses in-kernel device APIs
to issue writes and find out when they're on disk, the database ideally
uses O_DIRECT.

I looked around otn.oracle.com to find a doc that talks about
configuring and verifying AIO+O_DIRECT in the database but got tired of
searching.  You might be able to find something if you're more patient
than I was.

- z