Re: Oracle 10g RAC Question

Dusan Bolek wrote:

> "Howard J. Rogers" <howardjr_at_dizwell.com> wrote in message
> news:<4163091b$0$23894$afc38c87_at_news.optusnet.com.au>...

>> > More or less RAC is the beast formerly know as a parallel server. You
>> > are assuming right about disk devices. However, all nodes must see the
>> > same datafiles, so node 1 & node 2 & node 3 must see at least X. The
>> > storage config is the same for all nodes on RAC.
>> > Instead of raw devices you can also use a certified clustered
>> > filesystem.
>> > 
>> Where your sentence reads 'node 1, node 2 and node 3 must see at least
>> disk group X', that is therefore not generically true. It is only true
>> if, on those nodes, there are instances which seek to manage the same
>> database which happens to be stored on disk group X.
>> 
>> I also don't understand what "the storage config is the same for all
>> nodes on RAC", either. I think it falls into the same category of wobbly
>> distinction between RACing and Clustering.

But there is no such thing as a "RAC node", unless one adopts some very sloppy terminology (which Oracle Corporation is not immune from, I realise). There are nodes of the hardware cluster. And there are instances in a RAC.

Therefore, again, the sentence should be "participating instances in a RAC must all see at least one shared storage device".

Or something like that, anyway.

> All others could be assigned as
> needed for other purposes, but mount points for his clustered database
> must be accessible from all nodes.

From all *instances*.

I am not getting at you in particular. This is a general thing. Daniel does it. The Oracle documentation does it. All sorts of people do it. Even I have talked about having a 'two node RAC' in the past.

But there is a difference between instances and nodes. RAC clusters instances, not nodes. The language should reflect that fact.

Which is why I try to be very careful these days to talk about a 'two-instance RAC', for example.

> In his post he asked about three
> node RAC setup,

Meaningless, I'm afraid. He has a three node cluster. Or he has a three instance RAC. The distinction between the two is what I'm trying to highlight. And its a distinction which is important to grasp.

> so for me these nodes are the ones dedicated for RAC.
> We were not talking about general clustering, but about his planned
> RAC.
So let's all talk RAC terminology, then. Which, given that it is an Oracle product, means: let's talk instances, not nodes.

> The sentence about the same storage config should be taken from a
> database point of view. In his post, you can see the question if each
> node can have different tablespaces assigned (for several
> application), that is not possible and the *ORACLE* storage
> configuration (datafiles, tablespaces) must be the same for all nodes,
> which is obvious because RAC is only single database served by several
> instances.
> In this thread, we are not talking about general clustering.

I agree. So let's not use the word "node" when we mean "instance". Because otherwise, the confusion level rises.

That, generally, there is one instance per node should not mean we mix the terminology. The analogy is with the talk about instances versus databases in single-instance Oracle. A few years back, we used to get a lot of posts here about "shutting down a database", or "opening an instance". That seems to have died down a bit, because although there is a one-to-one correspondence between an instance and a database in a non-RAC/OPS environment, people finally seem to have realised that they are two different things.

I believe that realisation is a valuable thing, rather than just word games or nit-picking. Having a clear picture/definition of things is half the battle to understanding them.

Regards
HJR Received on Tue Oct 05 2004 - 05:14:24 CDT