setting up a vm-vm 2-node cluster in Oracle VM environment

[sborjigin]

I'm trying to set up a vm-vm 2-node cluster in Oracle VM environment using Oracle Sparc T3-2 server.

The two VMs (LDoms) are on the same physical server. Both nodes have SFHA installed, and ready for cluster configuration. But before I can do the cluster configurations, I have the following questions need to be answered:

1. How to set up private heartbeat links in this case?

2. The shared disk would be a netapp volume. Will it work?

3. Will I/O fencing work in this case? The fencing disks would be netapp volumes in small sizes that were presented to VMs.

Can anyone shed some lights on this?

Your help is greatly appreciated.

[mikebounds · Accepted solution]

Answers to questions:

1. Don't know much about LDOMs, but in VMWare you just create 2 virtual networks for the 2 heartbeats so I would assume you do the same if the 2 LDOMs are using the same phsical network cards.  If LDOMS are using separate physical neworks for heartbeats, then you need to connect these together as if they were two separate physical nodes
2. If NetApp volumes are presented as LUNS so they are seen in format, then you can use these as if they were normal SAN LUNs, using diskgroup and mount resources.  If they are presented over NFS, then this should still work and you just use Mount agent with FSType nfs
3. If NetApp volumes are presented as LUNS so they are seen in format AND they support SCSI III, then you can use I/O fencing, but I would doubt they will support SCSI III.

Why are you clustering within a single phyical node - if you loose the phyical node, then both VMs are dead, so not great for HA.  Also I/O fencing is a bit overkill for this set-up as you are unlikely to get split-brain betwwen 2 VMs in the same physical node.

Mike

[sborjigin]

thank you for taking time to go over my questions.

As for why we're clustering within a single physical node, this is a test environment and currently we have only one such hardware available for tests.

For heartbeat configs, I'll test with 2 pairs of virtual networks off the 2 physical network interfaces, and see how it'll work.

NetApp volumes are presented as LUNs, and seen in format. There was one thread about disks having to be only full SCSI disks in order to be under Veritas Volume Manger control, so I was just wondering if this applies to our case.

I'll update you if I run into any problems or have any findings on this issue.

SB

[mikebounds]

Disks only have to be basic SCSI to put under Veritas Volume Manager control AND VCS control, but if you want to use Fencing within VCS, then need to be SCSI III, but fencing is optional.

Mike

[joseph_dangelo]

The following guide will go over all of our support for LDoms/Oracle VM for SPARC.

http://www.symantec.com/business/support/resources/sites/BUSINESS/content/live/DOCUMENTATION/5000/DO...

Joe D

[TheNomad]

In the past, we have tried something similar with AIX running on VIO (virtual I/O) partitions and came to a rude awakening that, that configuration didn't support SCSI-III reservations. So, we wouldn;t be able to use fencing. Since it was an integral part of our testing strategy, we gave up clustering virtualized nodes, and went with AIX LPAR structure, where you can assign pieces of hardware (such as FC HBAs or NICs) to a partition and use no hardware abstraction layers in OS. So, depending on what you are trying to accomplish, you might be in the same situatin as I was, 4 years ago. I don't know much about Virtual Box partitions, let alone, them running on SunOS. So take it with a large grain of salt.