Re: clustering and high availability?

"Daniel Morgan" <damorgan_at_x.washington.edu> wrote in message news:1087195407.21563_at_yasure...

Well, it's a neat segue from defending the indefensible into a rather more important topic concerning what RAC actually is, but I'll run with it.

> You wrote: "Oracle doesn't define "shared everything"?"
>
> And Tom Kyte wrote this: "SQL Server has shared nothing clustering. we
> have and have had shared everything. With sql server, you put the data
> for department 10 on node 1, department 20 on the other node. If you
> need dept = 10 data, you goto node 1. need dept = 20, you goto node 2.
> With Oracle -- you goto the node you want, they all access any bit of
> data, no physical partitioning of data needed nor desired."
>
> Reference:
>

http://asktom.oracle.com/pls/ask/f?p=4950:8:12163950898199995897::NO::F4950_P8_DISPLAYID,F4950_P8_CRITERIA:3512484632553,
>
> Tom Kyte also wrote the following:
>
> "OPS is a shared EVERYTHING architecture."
>
> Reference:
>

http://asktom.oracle.com/pls/ask/f?p=4950:8:12163950898199995897::NO::F4950_P8_DISPLAYID,F4950_P8_CRITERIA:351213613475,
>
> Last time I looked Tom was a Vice President at Oracle Corporation and
> he seems to have no problem using the phrase "shared everything".

So? Because Tom says on one rare occasion that black is white, it is so? I said a few posts ago, that if Oracle did indeed claim RAC itself was a shared everything clustering topology, that would be wrong. I stand by that statement, because it would be to confuse clustering with RACing. They are two completely different topics. If it were true, how would it be possible to run RAC on a single laptop? Anyway, my reading of Tom's articles (which are not, by the way, the official Oracle documentation) is that he is using the term as convenient shorthand, to explain the difference between RAC and, for example, a federated database architecture.

But let's ask him: Tom, did you really mean to say that RAC was a clustering topology in its own right? Or were you using clustering topology language because RAC (usually) runs on top of such a topology, and therefore the two could at a stretch be encompassed with the same terminology? Perhaps Tom will help us out with a clarification. Then we can discuss whether he's right or wrong based on that clarification (and, incidentally Daniel, just because Tom is a VP at Oracle doesn't mean he turns water into wine, or falsities into facts... so cut the appeal to authority, because it won't wash).

> Neither does Kent Stroker Oracle's west coast US RAC expert. So given
> the choice of listening to another of your angry attempts to pick a
> fight ... or Tom Kyte's statements ... what do you think a reasonable
> person will do?

A reasonable person will ask themselves: is RAC actually clustering? Or do I have to accept the word of someone, who thinks he can magically intuit the motivations of other posters, that it is? When they accept that RAC is not itself a clustering technology, then they will readily accept that the use of clustering terminology to describe RAC is inappropriate and inaccurate.

Do I sound angry to you? Really? OK, in that case something has been lost in translation. Let me tell you how I actually feel. Bemused that you would arrogate to yourself the right to read my mind, know why I am posting, judge my sense of humour, believe yourself to be in agreement with Serge when he's actually posted he disagrees with you. Bemused isn't quite *a*mused, but it's certainly not angry either. So don't make claims you can have no possible knowledge about, OK?

> Last word is yours Howard as I'll not respond to this thread again.

Good. In that case, I'll summarise.

Serge thinks RAC is merely about sharing disks, nothing else: "I thought RAC is shared disk, not shared everything."
Daniel thinks Serge is wrong: "You thought wrong" I suggest Serge is right: "Therefore, Serge is I think correct to define carefully that RAC means shared disk storage, or shared database files, not actually shared "everything". "
Daniel thinks being precise on such a matter is a 'game': "If you are looking for an argument ... look elsewhere. I'll not play that game."
Serge agrees with me: "you're dead on what I meant." Daniel tries to avoid the issue: "[I'll agree with you and Serge] only if you agree to play nice with others for at least 15 minutes." Daniel thinks this is funny: "Get a sense of humor Howard. " Daniel tries to change the entire point of the thread "Serge and I were discussing hard disks."
And in this post, Daniel tries to change the entire point of the thread again: apparently, we're now discussing what Tom Kyte has called RAC in a couple of posts on asktom.com

I will merely conclude this thread by pointing out once again that RAC is an application. It is not a clustering technology, nor a clustering topology. It utilises and relies upon one particular clustering topology, and if you therefore say RAC is 'shared everything' I will know what you're talking about. But it is neither a clustering topology nor a clustering technology in its own right, and therefore actually, precisely, qualifies for no such description. Which is why I won't allow students or clients of mine to call it that without correction.

Is this wordplay or gamesmanship? No. The difference is significant, because far too many people are adopting RAC for completely the wrong reasons -and in large measure, because they don't truly understand what RAC is -and, perhaps more important, what RAC is not. It is therefore important to define, very clearly, what RAC is and isn't. And in that process, I have to say Tom's two posts quoted here are wrong, though again I know what he meant, and I think I know why he said it. I have no idea who Kent Stroker is, or what his opinions actually are (Daniel's precis of them doesn't count). But I doubt very much that Kent would argue seriously that RAC is a shared everything clustering technology in its own right... though maybe Kent can be persuaded to speak for himself on the matter.

Did I want a fight on this? Nope. I just wanted the facts about RAC to be free from the unfocussing effects of confusing RAC with clustering. Daniel doesn't seem to see the difference, which is his privilege and you can only take a horse to water, not make it drink -though it would be nice if the horse would acknowledge his reluctance to drink stems from his own wilful personality, not the externalities that have been invoked here, such as my wanting to pick a fight (false) or my lack of a sense of humour (ditto).

Whatever: hopefully, "reasonable people" reading this won't be prone to making the mistake of confusing clustering with RAC.

HJR Received on Mon Jun 14 2004 - 02:27:14 CDT