ESX 4 - Mirroring Datastore and Raw Disks

[ix2010]

Hello guys,

i have a question about HA

The environment:

• 2 x ESX 4 Server

• 2 x SAN Storage systems (IBM DS8000)

• SAN based mirroring is not available

We are using raw disks in physical mode for the VMs to enable SCSI commands. The mirroring of the VM data between the 2 SAN storage systems is performed in the VM itself with LVM mirroring on Linux.

So we can handle a failure of one DS8000 SAN storage system in the VM, because of the LVM mirroring of the data. But we still have the problem, that the datastore where the VM configuration (and raw disk mapping file) is so located is not mirrored. The datastore itself is also located on one of the SAN storage devices.

How can we mirror the VM configuration (and raw disk mapping file) in the environment from above?

Any suggestions are welcome ...

Thanks and Regards

[marcelo_soares]

For the OS disk you can create a datastore on the second DS8000 and attach a disk to it mirroring with LVM.

Mirroring the VMX files without storage mirroring should be done with a simple periodic copy. I am just in doubt if this will work in case of a failure, as your LUNs and VMDK's will not be available. You surely will need to shutdown the VM and reconfigure it in case of a failure...

Marcelo Soares

VMWare Certified Professional 310/410

Virtualization Tech Master

Globant Argentina

Consider awarding points for "helpful" and/or "correct" answers.

[ix2010]

Hello Marcelo,

thanks for your reply.

For the OS disk you can create a datastore on the second DS8000 and attach a disk to it mirroring with LVM.

I am sorry i was not clear enough.

All disk that are pass through to the VMs are physical raw disk and every VM has 1 OS disk on the the first DS8000 and 1 OS disk on the second DS8000.

The same for the data disks (1 data disk on the first DS8000 and 1 data disk on the second DS8000). So the VMs itself are fully mirrored by the Linux LVM in the VM.

The problem that i am facing is, that the OS disk for the ESX itself and the disk for the datastore (which contain the VM configuration and the raw disk mapping file) is also located on one of the DS8000 storage systems. So in case of a failure of one DS8000 one ESX server (and corresponding datastores) is gone.

The other ESX server has also mapped all LUNs which are the physical raw disk for the VMs, but the VM configuration files and the raw disk mapping files are gone, because of the datastore is not mirrored.

As far as i know it is not possible to mirror the datastore on ESX level (something like LVM mirroring or MD devices aka software raid).

In summary .. in case of a failure of one DS8000 i still have one of the physical raw disk of the VM, but no VM configuration file and raw disk mapping file.

The whole stuff should also work automatically (like in case of a failure of the ESX host itself) - no manual intervention in case of a DS8000 failure should be performed.

Thanks and Regards

[marcelo_soares]

Yeah... you will not have a way of doing that automatically by normal ways. Maybe some sort of scripting...

If you had san replication things would be easier.... In case of failure in this environment you will have to re-create your VMs and attach the to the LUNs replicated with LVMs. Is not so bad, at least your data will be there.

Marcelo Soares

VMWare Certified Professional 310/410

Virtualization Tech Master

Globant Argentina

Consider awarding points for "helpful" and/or "correct" answers.

[ix2010]

Hello Marcelo,

thanks again for your reply.

You are right, it would getting much easier with some kind of SAN replication, but unfortunately it is not available.

You are also right that no data is gone, but the environment should hit the HA requirements like minimal downtime with automatic fail over (something like HACMP on IBM AIX).

In some kind of way i also think that is is a big weak point of VMware that datastore mirroring is not possible (in general it should be possible technically).

Thanks for your reply .. if anybody else has implemented such a HA solution - please provide me some more information and possibilities.

Thanks and Regards

[AndreTheGiant]

Replication can be handled at different level:

• Storage / Array level -> require at least 2 storages with this function (on some products it could be also syncronous)

• host VMFS level -> for example with Double Take product

• host vmdk level -> with Vizioncore vReplicator or Veeam Backup

• guest block level -> like a Linux DRDB solution implemented between two guests

• guest volume level -> like LVM

• guest file level -> file DFS (v2 works also at block level), rsync, robocopy, ...

• guest application level -> for example with a MySQL cluster or an Exchange 2010 DAG

At which level you must work? It depends on what you want to do and your requirements.

Andre