[Ocfs2-users] Why OCFS2 with RAC

[David Johle]

I have been a user of OCFS2 for quite some time now (2 years or so) 
and a user of Oracle RAC for several years as well.  My usage of 
these two is completely independent though, the cluster filesystem is 
for application level usage (web servers, etc.) only because RAC can 
manage its own shared storage directly.  And that brings me to my 
observation/question...

I keep seeing a lot of messages on the list relating to the use of 
ocfs2 on systems running RAC--often with problems in the OCFS2 camp, 
but that's just the nature of this list.  What I'm trying to 
understand is why one would even consider adding OCFS2 to the systems 
running RAC.  To me that is just adding complexity to critical 
systems, and complexity tyipcally results in decreased 
reliability/availability in the long term.

I have iSCSI targets, available on multiple paths, presented as block 
devices with device-mapper-multipath.  My data & flash recovery are 
then managed by ASM, which directly uses those block devices.  For my 
OCR & voting disks, one could do the same with block devices, 
although I know there were some issues with using block devices on 
certain versions, but raw devices are another option there.  In my 
case, I simply use libraw to present the iSCSI based block devices as 
raw devices, and that's what they use.

Having RAC directly deal with the shared storage eliminates a lot of 
filesystem level overhead and removes a potential outside force that 
could unexpectedly bring DB nodes down (i.e. ocfs2 cluster stack 
fencing itself).  Not to mention the reduction in system 
administration resources for manging the filesystem.

One place I could see a use for the OCFS2 is a shared ORACLE_HOME 
among nodes, but that has its own pros & cons which can be debated on 
some other mailing list :)


So I'm curious, what benefits are there to having the OCFS2 available 
on the RAC system, moreso related to using it for CRS and DB storage purposes?