Re: Cluster file systems versus raw devices in Oracle RAC
From: Robert Blok <robert.blok@xxxxxxxxx>
To: oracle-l@xxxxxxxxxxxxx
Date: Thu, 29 Dec 2005 13:06:53 +0100
I am currently designing a similar system for a customer right now.
In my opinion using raw disks makes your architecture more rigid than
with a independant storage layer between the database and the disks.
This means that you have to solve more availability issues on a higher
level than I would like.
Providing your application layer (either databases or any other server)
with a storage platform (/oracle is available on all nodes), gives you
the flexibility you might want in an environment; every database can run
from any node.
In case of GPFS; version 2.3 is said to support a 2-node configuration
and no voting/quorum disks. I have been told that there is a tie-breaker
construction to have the two nodes watch eachother. I have no
information about directio yet (like in Polyserve ;-) .
ASM would be a solution between the storage and the nodes if only they
could storage both the Clusterware files and the Oracle software
inside... That we might need a third node in the cluster makes it a
complex solution which in my humble opinion is likely to errors...