Re: clustering and high availability?

"Daniel Morgan" <damorgan_at_x.washington.edu> wrote in message news:1086928161.576330_at_yasure...
> Howard J. Rogers wrote:

Steady on, Daniel...

> The phrase "Shared everything" has been very well defined by Oracle for
> many years.

Shared everything is a well known *clustering* phrase (as in, hardware clustering topology), and means shared *storage* with a DLM, even there. RAC certainly sits on top of a "shared everything" clustering topology, and will not work in a "shared nothing" clustering topology. But I don't think that makes RAC a "shared everything" clustering topology in its own right, particularly when RAC doesn't even need a cluster to work.

If Oracle documentation says RAC is itself shared everything, I'd say that was wrong. And that's without even starting to quibble about what the word "everything" really means.

>And as I know you are an expert I also know you know this
> without me having to tell you ... Something you clearly indicate in
> your response. And we both know you could have made mention of block
> sharing rather than RAM sharing but chose not to: Why?

Because Serge first mentioned "Shared everything, surely you mean shared disk". And you said, "No, shared everything". In other words, you were contrasting the word "everything" with Serge's use of "disk", and hence implying that something *more* than just the disk was being shared. Well, what else is there to share? RAM, CPU, maybe a graphics card??

I think my essential point (and I suspect Serge's too) is: Don't confuse (a) RAC's co-ordinated access across nodes to a single database with (b) hardware clustering. "Shared Everything" is a standard description of certain hardware cluster configurations. It is not one that can accurately be used to describe RAC.

> And even you Howard clearly read that the subject was related to
> disk. Neither Serge nor I said a single thing about memory. So you
> are, in my far less than humble opinion, trying to start something.

No I'm not. It was *you* that said to Serge "You thought wrong. Oracle has shared everything" -in response to Serge's perfectly accurate "shared everything in RAC only really means shared disk". So what did Serge 'think wrong' then?

By saying he was wrong, you were implying that rather more than disk access was being shared. Or so it seemed to me.

And as I also explained, the phrase "cache FUSION" can, and unfortunately by the uninitiated frequently is, also taken to mean that instances are magically merged across nodes; that somehow memory really is shared in a RAC. In a context where you say flat out that "shared disk" is "thinking wrong", and hence appear to be suggesting that something more than just hard disks are shared in a RAC, it would be a mistake to allow such a possible reading to stand uncontested.

> If you are looking for an argument ... look elsewhere. I'll not play
> that game.

Take a deep breath and read the gist of this thread again:

Serge: I think that in RAC "Shared Everything" really means "Shared Disk" You: You think wrong
Me: Actually he's technically accurate.

That's all. It's not a game. It's not an argument. It's not a put-down. It's a clarification.

Maybe you haven't had students who, when I explain that cache fusion is just a bunch of background processes passing messages across the network, say, "Oh. Is that all. I thought it was much more than that". But I have. So I'd rather the record was left crystal clear that it really is only the hard disk that is truly shared in RAC.

HJR Received on Fri Jun 11 2004 - 00:20:20 CDT