RE: Oracle cluster configuration : RH Cluster Suite mandatory

["Frank, Ryan" <Ryan Frank allvac com>]

I don't think I completely understand the comment "In a 2 node RAC, it
only takes the one node to fail to bring everything down".  I run
several 2 node RAC's using OCFS on RedHat AS, I have had machines fail
for many reasons, and never has it brought down the RAC?  In fact most
of our DBA's didn't even notice the other node was down, or we had it up
before they noticed.

The thing to consider is OCFS is strictly for Oracle.  When I say this,
I have been dealing with people writing other files to an OCFS volume
and causing OOM (Out Of Memory) kills.  Has to do with the
buffer_head_cache for the file system.  Oracle seems to feel that they
warn you not to put anything else on it, so they have not done anything
to resolve it... I am not sure if OCFS v2 has this issue, but I would be
pretty confident that GFS does not.

So I guess it would depend on what you intended to use the shared file
system for.  If you want it for general purpose and some Oracle RAC
stuff, then GFS may be the right choice to maintain consistency within
your environment.  I on the other hand only intend to use the shared
disks for Oracle, so I selected OCFS.

All-in-all I have been very happy with OCFS.  It has worked very well
with very little headache (Once we got rid of the OOM Kill problem).  

-Ryan

-----Original Message-----
From: taroon-list-bounces redhat com
[mailto:taroon-list-bounces redhat com] On Behalf Of Benji Spencer
Sent: Wednesday, August 24, 2005 11:12 AM
To: Discussion of Red Hat Enterprise Linux 3 (Taroon)
Subject: RE: Oracle cluster configuration : RH Cluster Suite mandatory

 From what I read in RedHat's FAQ on GFS (i think that is where it was),
if 
RAC is being used for availability (and not the pure need for the 
hoursepower), then you will want at least a 3 node RAC due to a process 
which runs on one of the GFS nodes. In a 2 node RAC, it only takes the
one 
node to fail to bring everything down. In a 3 node rac, two nodes come
down 
before everything is brought down. Donno why installing the "controller"
on 
each node isn't an option apart from overhead. If overhead is the issue,

then why would one want this to reside on any node?

Oracle claims OCFS is comparable to RAW devices. Donno where GFS stands
on 
this.

>What is Red Hat's stance of OCFS v2. vs. GFS?


---
Ben Spencer
Web Support
bspencer moody edu
x 2288

--
Taroon-list mailing list
Taroon-list redhat com
http://www.redhat.com/mailman/listinfo/taroon-list