[Info-vax] Hypervisors and clusters

[Grant Taylor]

On 1/22/21 6:11 PM, Mark Berryman wrote:
> I can't use a NAS because the only protocols on a NAS that present a raw 
> disk to the host are iSCSI and fibre channel.

My opinion is that both iSCSI and Fibre Channel are block based SAN. 
Conversely, NAS is file based.  There may be some things that call 
themselves a NAS that actually provide SAN protocols.

> If I try to run the iSCSI initiator on a current version VMS, it 
> crashes the system.

:-(

> If there is a NAS out there that offers fibre channel support usable 
> by VMS, I haven't found it.

I don't know about a NAS per se, but I suspect there are a lot of 
systems that can do it.

I wonder how well Linux I/O SCSI targets might do with a supported Fibre 
Channel HBA exporting LUNs to VMS.

> My cluster is currently SAN-based.  That means that all disks are named 
> $1$DGAn:  That means that, in order for a virtual host to join this 
> cluster and have direct access to the storage, the virtual host has to 
> think it is talking to an HBA.  It can't be something mapped to a disk 
> available to the host system.

Why is that?  Does such a mapped disk show up as a different device 
naming scheme?  (Obviously I don't know much about VMS and SANs.)

> Are there any hypervisors out there that will pass access to a local 
> HBA directly to the guest host?

Yes.

There is a technology that is commonly known as "PCI Passthrough". 
(Don't worry about PCI / X / e nomenclature.)

I've used PCI Passthrough on ESX(i) and VirtualBox.  I'd be surprised if 
other contemporary hypervisors didn't support it in one form or another.

VirtualBox does have an option to use iSCSI to back a virtual disk. 
It's annoying to do and can't be done through the GUI.  I think that 
VMware has similar functionality.  --  Though this may still suffer the 
drive name issue that you're alluding to above.

> If so, does the guest host need a driver for that specific HBA or 
> does the hypervisor map it somehow?

Yes.  The guest OS does need a driver for the card passed through the 
hypervisor into the guest OS.

> My specific situation:
> 
> All of my non-VMS systems are Macs.  I have no Windows, Linux, or other 
> systems.  The HBA on the Macs is plugged into a thunderbolt port.  Now, 
> since I doubt thunderbolt support is being added to VMS as part of the 
> port, I'm wondering if a virtual host is going to work for me.

Do you have something like a PCIe / PCI-X / PCI adapter that connects to 
the Mac via Thunderbolt?  If so, you might be able to pass one of the 
PCI* slots through to the guest VM.



-- 
Grant. . . .
unix || die