Re: RAC node distance

From: Steve Baldwin <stbaldwin_at_multiservice.com>
Date: Thu, 25 Mar 2010 15:08:46 +1100
Message-ID: <d282b3ab1003242108n5b9ac7b7v70300d2938ded6e7_at_mail.gmail.com>



Hi Ram,

I have no experience with 'geo-clusters' but we do use RAC so for what it's worth ...

In my opinion this is a bad idea. No advantages, only disadvantages. Keep in mind that each cluster node needs at least 3 separate network connections - one for the interconnect, one for shared storage (assuming you are using iSCSI - don't even know *if* you could do it with fiber channel) and one for the client network. Having 3 (min - ideally more for redundancy) separate high speed (min 1 Gb, ideally higher) networks spanning 3 city blocks sounds expensive and doesn't buy you anything anyway. Where are you putting the shared storage - in a third building? If they think this offers some kind of DR solution they are sadly mistaken. A DR solution needs to duplicate *everything*. Servers, shared storage, network infrastructure, ... Putting a single RAC node 3 blocks away is not something I would consider (unless you can give us more insight into their reasoning).

HTH, Steve

On Thu, Mar 25, 2010 at 2:20 PM, Ram Raman <veeeraman_at_gmail.com> wrote:
> Hi
>
> We have a 2 node RAC for one of our new applications. There are 2 sets
> of application/web servers; each set located on different building separated
> by about 3 city blocks. But, both the RAC nodes are in one of the buildings.
> The management seemed to be inclined to want to separate them in different
> buildings. To me that does not sound like a great idea, with interconnect
> traffic and such. Questions:
>
> 1) Are there any sites that use RAC 'geo-cluster' mode?
>
> 2) Would running the RAC in 'active-passive' mode help in case of geo
> cluster type solution? What kind of interconnect is used by sites that have
> geo cluster type RACs?
>
> PS. I am aware that RAC is not a DR solution, going for a dataguard would be
> a good DR solution.
>
> Ram.
>
>

--
http://www.freelists.org/webpage/oracle-l
Received on Wed Mar 24 2010 - 23:08:46 CDT

Original text of this message