On using a single machine as the failover for two

[Vishal Augustine]

Hi,

Is it possible to configure ServiceGuard for three machines - A, B and C - such that C works as the failover machine for A as well as B.

1. If A fails C will comeup for A. After A's failure, if B fails C won't take up any corrective actions
2. If B fails C will comeup for B. After B's failure, if A fails C won't take up any corrective actions


Is it possible to have such a triangular arrangement ?

Thanks and Regards
Vishal

[Bill Hassell · Accepted solution]

Yes, you can have a cluster of 3 machines. However, you should not think of machines as the switchover point, but packages instead. Now if you only have one package between the 3 machines, then your scenario is reasonable. But if you think of packages located on a specific machine, the transfer of the package can be explicitly defined depending on the states of each machine.

For instance, you can run 4-5 packages on the 3 machines at the same time and a failure on one machine would follow each package's definition as to where it should move.

[Vishal Augustine]

I am sorry. When I mentioned failover machine I assumed tht all the packages (x,y,z) are running on A and B. C is configured to bring up those packages (x,y,z) when either A or B is down due to some issue like network card problem.

Thanks Bill

Regards
Vishal

[Nick Wickens]

If you have three similar machines and three applications you could have one application running on each node by setting the node as the prmary node for each package. Would mean that you are then spreading your work load across all three machines for normal processing.

[Trond Haugen]

"Anything is possible."
It is all in the package scripts to configure how they start up and where they can switch to.
Another question will be if all machines have "power" enough to run as many packages as you want.

Regards,
Trond

[John C. Shoemaker]

As a side topic, do all of these machines have unique sources of UPS regulated building power; phone lines etc?

[Misa]

There's an HP product you might want to look into for your configuration if performance will be an issue.

The product is called PRM (Process Resource Manager, I think). It's supposed to limit the amount of resources that can be consumed by each application so that overall performance is more predictable/within bounds.

I've never had occasion to do it myself, but it came highly recommended. I suspect it's not simple to implement, however.

--Misa