Re: RAC and scalability

From: <fitzjarrell_at_cox.net>
Date: Mon, 7 Apr 2008 08:02:03 -0700 (PDT)
Message-ID: <bb115036-bb3a-40f5-93e8-9c61e8e597ad@a70g2000hsh.googlegroups.com>


On Apr 7, 9:50 am, Andrea <netsecur..._at_tiscali.it> wrote:
> On Apr 7, 3:53 pm, Mark D Powell <Mark.Pow..._at_eds.com> wrote:
>
>
>
> > Andrew, I know nothing about the hardware you listed but if you want
> > useful responses you need to identify the type of application that you
> > will be running: OLTP, DSS, OLAP, etc ... and identify the current
> > user load.  The Oracle version you will be using is also important.
>
> > For the solution to be RAC it must involve at least two nodes.
>
> > Also depending on the application design some applications just do not
> > scale well with RAC.  To really run well under RAC the application
> > should have been designed with RAC in mind.
>
> unfortunately i can't collect informations on current user loads,
> because this is a new project that merge several applications and
> these apps doesn't current using.
> The type of applications will mainly OLTP on 10g rel. of rdbms.
> Sure 4 CPU is better then 2CPU, but what i want to know is if ScaleUp
> is better then ScaleOut (horizontal scaling) in reference to cost of
> the 2 servers describes up.
>
> So if bl480c server with dual CPU have cost of 6.000$ and bl680c with
> 4 CPU is 12.500$ (double), in RAC environment it is better 2 nodes of
> bl680c or 3-4 nodes of bl480c ?
>
> thanks again
> Andrew

Who could say with that limited bit of information? How much RAM is installed in each configuration? How many concurrently connected users do you expect? How fast is your network? What size do you expect for the SGA?

With two nodes you have limited options for TAF (obviously) and should one node die you're left with a single point of failure. With four nodes you have a bit more 'wiggle room' before you end up with that situation (three nodes must fail, and I expect that is a much rarer occurrence).

This involves far more than simple monetary costs; additional 'expenses' for this are availability, scalability and performance, and you might find that the four node configuration better suits your expected user load and your anticipated SLA. But, no one can say anything for certain until you provide more information.

David Fitzjarrell Received on Mon Apr 07 2008 - 10:02:03 CDT

Original text of this message