Re: RAC design question

  • From: Martin Bach <development@xxxxxxxxxxxxxxxxx>
  • To: Jed_Walker@xxxxxxxxxxxxxxxxx
  • Date: Tue, 16 Aug 2011 11:52:19 +0100

Hi Jed,
indeed, the service will fail over in case your preferred instance crashes. However, your sessions won't-it's quite simple to test. Define TAF at the service level, define 1 preferred and n (where n > 0) available instances, start the service, connect to the service and then kill the instance. You can do this in SQLPlus, and you'll see that your connection lost contact.
Does that quick reply make sense? I don't have a system available right now, but can do a test if you like and share the output. 

Best regards,

Martin

On 15/08/2011 16:04, Walker, Jed S wrote:


Thank you Martin, Frits, Toon, and Kumar.
Question though, If I have a service with one preferred instance and multiple available instances I believe it should still failover with TAF to any of the available instances shouldn’t it? My understanding is that multiple preferred instances is for spreading the load across multiple instances, not failover. 

*From:*Martin Bach [mailto:development@xxxxxxxxxxxxxxxxx]
*Sent:* Saturday, August 06, 2011 7:33 AM
*To:* Frits Hoogland; toon.koppelaars@xxxxxxxxxxx
*Cc:* Walker, Jed S; oracle-l@xxxxxxxxxxxxx
*Subject:* Re: RAC design question

Hi,

I would like to put one or two more points into the discussion.
If you would like to prevent connections from dropping by employing TAF you need at least 2 preferred instances. I also got best results from FCF with the same setup, plus it could give you runtime load balancing. But then again I haven't heard of anyone using FCF (and UCP) in real world applications ...
Since you didn't tell us more about your application you need to decide if these points are applicable.
If you really only needed higher availability you could have with an active passive cluster and saved on licenses... 

How this helps,

Martin

Martin Bach

Martin Bach Consulting
http://martincarstenbach.wordpress.com
http://www.linkedin.com/in/martincarstenbach

----- Reply message -----
From: "Frits Hoogland" <frits.hoogland@xxxxxxxxx>
Date: Sat, Aug 6, 2011 08:57
Subject: RAC design question
To: "toon.koppelaars@xxxxxxxxxxx" <toon.koppelaars@xxxxxxxxxxx>
Cc: "Jed_Walker@xxxxxxxxxxxxxxxxx" <Jed_Walker@xxxxxxxxxxxxxxxxx>, "oracle-l@xxxxxxxxxxxxx" <oracle-l@xxxxxxxxxxxxx>
With the clusterware you can setup a service for every schema which can fail 
over to another instance.

Frits Hoogland

http://fritshoogland.wordpress.com
mailto:frits.hoogland@xxxxxxxxx <frits.hoogland@xxxxxxxxx <mailto:frits.hoogland@xxxxxxxxx>> 
cell: +31 6 53569942
Op 6 aug. 2011 om 08:22 heeft Toon Koppelaars <toon.koppelaars@xxxxxxxxxxx <mailto:toon.koppelaars@xxxxxxxxxxx>> 
het volgende geschreven:

I think you've answered that design question very wisely.
On Sat, Aug 6, 2011 at 12:49 AM, Walker, Jed S <Jed_Walker@xxxxxxxxxxxxxxxxx 
<mailto:Jed_Walker@xxxxxxxxxxxxxxxxx%0b>> wrote:

>  Hi,****
>
> ** **
>
> I’m new to RAC, but have a question. We have a 5 node RAC that supports
> multiple markets each of which has its own schema. Due to each market having > its own schema, there is no sharing of blocks between markets. As such, I am > thinking that it would make sense to have each market work on only one node > because that would avoid having blocks passed between nodes, and thus should 
> be good for performance. (Note: the intent behind RAC was for high
> availability, not for scaling, each node can handle the workload of multiple 
> markets).****
>
> ** **
>
> Thoughts?****
>
> ** **
>
> **-          **Jed****
>
> ** **
>



--
Toon Koppelaars
RuleGen BV
Toon.Koppelaars@xxxxxxxxxxx <mailto:Toon.Koppelaars@xxxxxxxxxxx>
www.RuleGen.com <http://www.RuleGen.com>
TheHelsinkiDeclaration.blogspot.com

(co)Author: "Applied Mathematics for Database Professionals"
www.rulegen.com/am4dp-backcover-text <http://www.rulegen.com/am4dp-backcover-text>

Other related posts:

  1. Home
  2. oracle-l
  3. Archive
  4. 08-2011