RE: Question about RAC

  • From: "Michael Erwin" <michael.erwin@xxxxxxxxxx>
  • To: <ax.mount@xxxxxxxxx>, <oracle-l@xxxxxxxxxxxxx>
  • Date: Mon, 30 Oct 2006 13:31:01 -0600

Hi Alex,

 

Logical or Physical partitioning (L-Pars) a single host environment to support 
multiple RAC instances is logical as long as the host server can provide an 
extreme fault tolerance they need for HA, specifically think Sun 25k, HP 
SuperDome, IBM p590 class systems. This strategy allows customers to grow & 
migrate slightly easier than building a 2 node, then adding 2 nodes in the 
future. In these environments, each partition acts & is managed as you would 
multiple hosts.

 

Many times this is the case with customers that outsourced IT environment 
management, i.e. Perot, EDS, IBM, where the customer pays for physical system, 
not L-Pars. In the past I have seen customers buy the biggest boxes on the 
planet, and then l-par them into 2 & 4 CPU host boxes since it saved them on 
the outsourced support, in effect spending $5million, so they wouldn't have to 
spend an additional $300k per month to support 6 smaller hosts. Then they could 
change the partition configuration as needed or warranted.

 

Having said all of that, what is still illogical is for customers to put 
multiple related instances on the same physical partition of hosts, i.e. 
Reporting/OLAP Instance along with the OLTP Instance on the same physical 
CPU/Memory/Bus partition or host.

 

Hopefully, your customer isn't considering the use of VMware or Parallel to 
partition a single small i86 host for production environments.

 

 
Michael Erwin
Hotsos Enterprises, Ltd.
http://www.hotsos.com
Hotsos Symposium 2007 / March 4-8
Visit www.hotsos.com for curriculum and schedule details...

________________________________

From: oracle-l-bounce@xxxxxxxxxxxxx on behalf of amonte
Sent: Mon 10/30/2006 1:24 PM
To: oracle-l@xxxxxxxxxxxxx
Subject: Question about RAC


Hi all
 
I have a question about RAC.
 
I have implemented a few RAC on a couple of customers. I have always worked 
with two nodes and a single database, i.e 2 servers and a common database in a 
shared storage.
 
I have a customer who wants to use 2 servers to implement 4 RACs, i.e hacing 4 
instances in each server. Has anyone done this sort of implemntation? I am not 
sure if this is logical or ilogical :-)
 
 
Thanks
 
Alex
 

Other related posts: