Re: Oracle RAC for scalability or High Availability only

dananrg_at_yahoo.com wrote:
> I read Mogen's white paper. Thanks for the suggestion.
>
> Is RAC scalability typically only justifiable when you need to purchase
> a server that has more than 4 CPUs? Someone mentioned a 10 CPU box. I
> don't have that magnitude of growth to worry about, and I don't need to
> have HA.
>
> In my case, I am looking at two scenarios:
>
> 1) A small RAC configuration, we'd have two-nodes, each of which would
> be a dual-CPU machine.
>
> 2) In the second scenario, we'd have a single quad-CPU-capable SMP
> server (that we cannot grow beyond 4 CPUs), purchasing it with only 2
> processors at first, then scaling vertically by adding another 2
> processors when needed, plus additional memory. Although testing may
> show we'll need all 4 CPUs out of the gate.
>
> At what rate of growth would RAC make sense financially from a
> scalability-only perspective? Let's suppose that in 3-5 years the
> system is tapped out and our user base and/or system resource
> requirements doubles. In that time, if Moore's Law is still holding-up,
> wouldn't it be cheaper overall to replace the single SMP server with
> the latest and greatest hardware rather than use RAC? Yeah, I know it's
> a pain to migrate to another server vs. simply adding another node. But
> at what level of forecasted scalability does RAC become a must (or
> highly recommended) vs. a single SMP server? 50% growth? 100%? 200%?
> 400%?

Kudos to Daniel for his post about the reasons to use RAC.

Over the last decade, I've seen several sites of differing sizes do the upsize single hpux server route. It has worked well for them, often meaning the older server can be relegated to standby or development status. I don't really know how linux does on hp servers, I've become an hp-ux bigot because it runs so fast on the 9000/800 hardware (unfortunately, my linux experience is limited to the 2 dozen or so I have at home, not the modern enterprise versions, some of which are sitting in my to-do pile with certain books). As the stock boilerplate says, what was true in the past may not be true in the future.

>
> I'd like to hear more arguments in favor of a single SMP server, just
> to know RAC's proper place based on expected growth of users and/or
> system resources. Thanks muchly.

My gut feel is in the hundreds of %, but there is a lot of salt to be taken there. Even if true, it may only be true for an Oracle version or two. My general feeling is, if a place isn't willing to explicitly support what is necessary to be bleeding edge, let someone else bleed. Of course, I'm nostalgic and envious of those that get to do it and be paid for it. I just know better than to fight reality in certain places. As in so many of these types of decisions, in the end the technical issue is often the least part of it.

jg

--
@home.com is bogus.  "I cut to the chase - what I would want to
give and what I would want to receive.  A guy wants to say he still
loves her. A gal wants to know he still does. She wants to get goose
bumps. He wants to think he'll get lucky."  - Marcia Meulengracht,
author of last years most popular Hallmark Valentine's day card.