Jimbo wrote:RAC is not Cloud. Despite what the marketing suits may spin.
I recall challenging Oracle on their Grid concept (I suppose really now replaced with their cloud concept ) which was lauded as being able to add in any spare servers to consolidate processing power. I then asked the question could you therefore combine a Windows Server with unix server in regards to the processor grid. In pure grid theory the answer should be yes but when pressed Oracle they admitted no.
Q3. What about binary compatible nodes eg one node being RHEL and the other being OEL ?Yes. In fact did this recently on a test cluster. Not a good idea though (obviously) as there can be some kernel driver issues (e.g. ADVM of 126.96.36.199 runs on 5.4 kernel and not on the 5.7 kernel). Also makes o/s management more complex. I'm a big believer that each and every server node in a Grid/RAC needs to be identical at o/s level. Same kernel version, same drivers, same rpm's, same +/etc+ configs (such as hosts, multipath, udev) and so on.
Q4. Finally what about 2 nodes that were the same OS but had a slight different version of ORACLE_HOME eg 11gR1 and 11gR2 ?Well, you can have multiple Oracle Homes on the same server. So you should be able to have a 10 node Grid, policy managed RAC, server pool 1 running a 10gr2 RAC database as 5 instances, and server pool 2 running a 11gr2 RAC database as 5 instances. Offhand I do not see why this would not be possible.
any thoughts appreciated,Well, the obvious thought is why do you want to throw heterogeneous stuff together into an Oracle RAC? To achieve what need, and address which requirement, exactly?
KODS wrote:Correct - but these checks (like the same o/s kernel version for Linux) can be circumvented by setting env variable IGNORE_PREADDNODE_CHECKS to "N".
in RAC the utlity CLUVFY is verifying the node equavalence before installing the cluster software.
Even while adding the new node to the cluster, the pre-requisite is to run CLUVFY utility.