One Node pushing a RG off another node
We are discussing the following scenario:
2 Node Sun Cluster 3.1 each 24 GB Memory
Node A
rg1, Oracle Instance 1: 15 GB SGA
Node B
rg2, Oracle Instance 2:2 GB SGA
rg3, Oracle Instance 3:2 GB SGA
rg4, Oracle Instance 4: 10 GB SGA
So we have not enough memory on the nodes to run all resource groups on the same node at the same time. As instance 4 is less important, the idea is now to stop it in case one node fails, i.e it should not switch to node A if node B fails, and if Instance 1 switches to node B because node A failed, it should go off line.
Part 1 seems to be easy, just don't give it node a as potential primary. But for the other case, I don't know how to force it to go off line. I assume if i define a strong negative affinity between rg1 and rg4, rg1 will just simply not switch to node B instead of pushing out rg4.
BTW, I hope it remains a theoretical discussion, I know it is a stupid solution, it is not worthwhile to save the cost of the additional memory compared to the long term risk of such a solution, but some people want to know if this is possible or not.
Thanks
Fritz
