Re: Thoughts on instance crash and RAC

On Wednesday, August 21, 2013 8:46:15 AM UTC-7, Mladen Gogala wrote:
> On Tue, 20 Aug 2013 10:51:02 -0700, Mark D Powell wrote:
>
>
>
> > An alternate to RAC is buy a big enough machine to support your user
>
> load and then run Data Guard to provide a failover instance.
>
>
>
> Another alternative is to buy a NUMA machine, like SGI ALTIX, SUN T4-4,
>
> HP SuperDome or IBM xSeries. NUMA boxes usually come in as blades, which
>
> form a single huge server. Oracle is NUMA aware and can survive a blade
>
> failure, which is why people are buying RAC.
>
>
>
>
>
>
>
> --
>
> Mladen Gogala
>
> The Oracle Whisperer
>
> http://mgogala.byethost5.com

Hi Mladen my old friend! Please allow me to comment on your comment. All multi-socket systems are NUMA at this point. Nobody remains to make a FSB server. SGI's flagship is now called UV and it scales to 256 sockets and 64TB. It boots RH and SuSE from the stock iso! Oracle have moved beyod T4 to T5 and T5 scales to 8 sockets with a single hop. The Sun X3-8 (aka x4800) is a E7 based 8 socket box (2 hop). As always xSeries has the best commodity-CPU based NUMA because deep down inside is the remnant of Sequent :-). HP still sells the 980 (PRIMA) as well.

We agree though. I find it utterly absurd that anyone would use RAC cache fusion to scale multiple little boxes when it is so simple to buy a larger NUMA system an run scalable SMP. SMP (even a poorly made NUMA SMP like the Sun X3-8) will scale better than multiple boxes stitched together with RAC.

PS. For history sake let's recall that the SuperDome started life as the Convex Exemplar which was a really good server. Memories. Received on Sat Sep 14 2013 - 17:14:55 CEST