Re: clustering and high availability?

"Daniel Morgan" <damorgan_at_x.washington.edu> wrote in message news:1086914524.786060_at_yasure...
> Serge Rielau wrote:
>
> > I thought RAC is shared disk, not shared everything.
> > Cheers
> > Serge
>
> You thought wrong. Oracle has had a shared everything architecture
> for a very long time.
>
> Anytime you want to see 8 separate instances on 8 separate nodes
> simultaneously write to a single datafile come to Seattle.

Shared *everything*??

So node 1 is able to directly access and write to node 2's RAM? Or node 7 is able to process instructions on node 4's CPU?

Cache fusion might be clever, but it's not *that* clever!

The usual definition of "shared everything" in clustering circles is something along the lines of "In the shared-everything cluster model, all servers or cluster nodes in the cluster are given equal to the shared resources". Now, even granted that we are talking Oracle software here, and not physical hardware, the 'shared resources' in a RAC are, and are only, the database files. Not, for example, the instances. The instance on Node 1 cannot directly access and write to buffers in Node 2's buffer cache, but has to ask politely for the data to be shipped to it so that it can work on it 'locally'.

Therefore, Serge is I think correct to define carefully that RAC means shared disk storage, or shared database files, not actually shared "everything". Ordinarily, it would be a bit of a joke and technical shorthand, of course. But since Oracle decided to make up and sell RAC on the basis of the 'cache fusion' marketing hype, which does indeed suggest to the unwary that, somehow, instances really are 'merged' or 'fused' across nodes, I think Serge's point is actually not a bad one to make now and again.

Regards
HJR Received on Thu Jun 10 2004 - 20:14:30 CDT