Originally posted by: SystemAdmin

Hi all,

I am about to verify correct failover on a newly installed hacmp cluster.

When it comes to network failures, on a node with dedicated adapters, I would disable the switchport.

How would you simulate a network failure with virtual ethernet?

Cheers
delgado

Originally posted by: Casey_B

Hello Delgado,
I haven't set up a cluster with vio in a production
setting.

Here is what I am thinking, though....

When you have a normal network with an adapter,
you test by disabling the port. This simulates
either a cable, or switch failure.

These are two of the networking events that you
would like PowerHA to protect against.

Now, with a vio setup, you need to consider the same
thing. What events do you want HA to protect against?

The top two on my list would be VIO server failure,
and cable, or switch failure.

So, if possible, I would personally test by disabling the
ports associated with the vio server first.

Then as a different test, maybe trying to make the vio server
unavailable. For instance by rebooting the vio server.
(Haven't done this myself, I assume it is possible)

Both of these are of course not possible if there are other clients
attached to the vio server.

Two APAR's that you need to be aware of with HA and vio are the following:
IZ01331: NEW NETMON FUNCTIONALITY TO SUPPORT HACMP ON VIO
IZ01874: CLARIFY HOW TO CHOOSE IP ADDRESSES FOR NETMON.CF
Not only do you need to have them installed, but you need to also
read and understand the instructions provided.

Hope this helps,
Casey

Originally posted by: SystemAdmin

Very similar to disabling the ports on the dedicated adapters, depending on the type of VIO network i.e., SEA failover or NIB on clinet, you could reboot one vio server at a time and validate that the HACMP network is functional. If the Dual VIO server configuration has been validated, then rebooting VIO servers would simulate the failure scenario. I am assuming that the testing VIO servers would done that would involve diabling the port on the switch or on the server.

Thanks
Nagesh Potluri
IBM Service Delivery Subject Matter Expert
Technical Webmaster
Email:nbpotlur@us.ibm.com
Phone: 678-807-2290
Text Message: 6784587809@vtext.com

Originally posted by: SystemAdmin

Hi Delgado,
As the previous posts have said, the testing should focus on the loss of the VIO server. Ensuring that the loss of a VIO server doesn't cause the network to become unavailable on a node that is using it should be part of the HACMP implementation. Although this doesn't involve configuring any part of the HACMP product itself, eliminating the VIO server as a single point of failure is the goal. There are two ways to do this, both involve a dual VIO configuration. In both cases, there will be a single adapter in the HACMP network (a single virtual ethernet adapter for SEA failover, or an etherchannel NIB device that covers two virtual ethernet adapters). Shutting down one of the VIO servers at a time should test that the loss of a VIO server will not cause the loss of network activity on the HACMP lpar. No HACMP processing should occur.

To be more thorough in your testing, you could do as the others have suggested and disable network connections to the VIO servers in different ways. Again you would be verifying that network traffic is not interrupted on the HACMP lpars. This is technically only testing the dual VIO server setup, not the HACMP setup, but it is very important, nonetheless.

For more information on the dual VIO setup, see the Redbooks:
PowerVM Virtualization on IBM System p: Introduction and Configuration (SG23-7940-04), the chapter on Advanced VIO configurations
and
PowerHA for AIX Cookbook (SG24-7739-00)

I hope this helps,
Bill Miller
IBM Training