Re: Oracle RAC Performance - Two node test provides scaling to 10+nodes?

"Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote:
>
> As Gopalakrishnan points out, the critical limit it three
> nodes. In fact you get a clue from the 10g wait events
> which list : '2 way cache fusion' and '3 way cache fusion'
> and don't go above that.
>
> The worst case scenario is:
> Node A says to Node B
> I want block X, you are the directory controller for X
> Node B says to node C
> You have block X, give it to node A
> Node C says to node A
> Here is the block
>
> Asynchronously, C and A confirm to X that
> the block has been sent and received.

I'm not sure I buy this argument. It seems to me the worse case scenario is more like:

Nodes A, D, J, L, M, O, Q, R, T, U, V, X, and Z all independently say to Node B "I want block X".

And then node B says "Node C, give X to A, and the rest of y'all sod off for now".

I can see that if something scales perfectly (i.e. no contention) to 3 nodes, then it would probably also scale to many nodes. But if it merely scales well (i.e. an acceptably low amount of contention) to 3 nodes, I don't see that it will necessarily continue scaling well to many nodes.

> One warning - someone else made the comment that
> the manuals say "if is scales on one node it will scale
> on RAC". What the manuals actually says is "if it
> won't scale on one node it won't scale on RAC".

Xho

-- 
-------------------- http://NewsReader.Com/ --------------------
Usenet Newsgroup Service                        $9.95/Month 30GB