Shared SCSI Volume

[Bjoern_Gies]

Hello VMTN community,

I got two SuperMicro servers, each with an Adaptec 29320ALP-R[1] U320 parallel SCSI controller connected to the same external RAID system.

The RAID system is an EUROstor ES-6600[2] with two U320 channels.

I'm running ESXi 3.5 on both systems, seperated to each other.

The EUROstor RAID system offers the possibility to propagate a RAID volume to both channels - for clustering as they say.

I wonder if the ESXi can handle such a shared volume or if there will be problems when both are accessing the volume. I thought about having some virtual machines stored on that volume but only registered/mounted in one inventory of the two ESXi at a time to reduce possible conflicts. So that I can easily run the virtual machine on the other host, maybe when one physical server fails.

To illustrate it a bit:

RAID system[/u]

   [u]System-Volume-1[/u]

      Host1 only (SCSI Channel 0; ID 0; LUN 0)

   [u]System-Volume-2[/u]

      Host 2 only (SCSI Channel 1; ID 0; LUN 0)

   [u]VM-Volume-1[/u]

      Host1 only (SCSI Channel 0; ID 1; LUN 0)

   [u]VM-Volume-2[/u]

      Host2 only (SCSI Channel 1; ID 1; LUN 0)

   [u]Shared-Volume[/u]

      Host1 & Host2 (SCSI Channel 1&2; ID 2; LUN 0)

I found alot about shared SCSI disks for virtual machines, but not for ESX(i) itself. Found something about "cluster aware" OS, but most of them talked about SAN.

Thanks for any help or tipps, links whatever that might help me on that topic in advance!

Greetings from Germany

Björn

http://www.adaptec.com/en-US/products/Controllers/Hardware/scsi/entry/ASC-29320ALP-R/

http://eurostor.de/german/ES6600U.D.php

[DSTAVERT]

It is possible. Some cluster type arrays by HP and Dell do support it. VMFS, the ESX(i) filesystem, is a cluster type filesystem. ESX(i) will keep you from starting a machine on two hosts simultaneously. If you have had success please do a more thorough report. I am sure there would be interest.

[Bjoern_Gies]

Hi DSTAVERT,

thanks, so it's worth a try.

Gonna check that next week.

Greetings

Björn

[Texiwill]

Hello,

Unless the Array has a SAN like controller ala HP MSA 500, then this type of configuration will not work as there is nothing controlling the locks required by ESX. Dell does have a similar device.

However, 'Disk Trays' will not work. They work for MSCS as MSCS handles the lock 'communication' via a private network.

Best regards,
Edward L. Haletky VMware Communities User Moderator, VMware vExpert 2009

Virtualization Practice Analyst[/url]
Now Available: 'VMware vSphere(TM) and Virtual Infrastructure Security'[/url]
Also available 'VMWare ESX Server in the Enterprise'[/url]
[url=http://www.astroarch.com/wiki/index.php/Blog_Roll]SearchVMware Pro[/url]|Blue Gears[/url]|Top Virtualization Security Links[/url]|
[url=http://www.astroarch.com/wiki/index.php/Virtualization_Security_Round_Table_Podcast]Virtualization Security Round Table Podcast[/url]

[savantsingh · Accepted solution]

The ESX would add a lock to the VM when it is in use. Regarding the "cluster aware" volumes you are not running a Virtual Center server so ESX hosts are not in a cluster. In other words the storage is not shared on a VC cluster level.

Having said that i agree with Texiwill unless the array controller supports shared LUNs it wouldn't work.

But it will nice to know the results if you can test sharing the volume datastore (shared volume) on both hosts. I would assume if you try and power on the VM already in use on the other host you would get an error saying " cannot open file or something" cause of the lock.

If you found this information useful, please consider awarding points for "Correct" or "Helpful".

[DSTAVERT]

Björn

I took it that you had already done it. As Edward says unless the array device has SAN type controller you would be out of luck. If you use the array attached to another server and added a software package to provide either iSCSI or NFS you would be able to share the storage.

[Bjoern_Gies]

Hello all together,

thanks for all your answers. I gave it a try this morning.

A quick overview what I've done and the results:

- Booting both servers with the shared volume [ [u]Success:[/u] Servers are booting ]

- Creating VMFS on the shared volume through Host#1 via VI Client [ [u]Success:[/u] Datastore added ]

- LVM.EnableResignature set to '1' on Host#2 [ [u]Success:[/u] Just created VMFS showing up ]

- Creating new VM on the shared volume through Host#1 [ [u]Success:[/u] VM created and booting (up to BIOS - no OS installed) ]

- Registering the VM on Host#2 [ [u]Success:[/u] VM registered - shown as powered off ]

- Powering on the VM through Host#2 [ [u]Failed:[/u] "Could not power on VM : No swap file." \ "Failed to power on VM" ]

- Powering off the VM through Host#1 [ [u]Success[/u] ]

- Starting the VM through Host#2 again [ [u]Success:[/u] "Virtual Machine Message" \ "msg.uuid.moved"(1) -> Choosing Keep, Booting afterwards ]

Not looking that bad - for a simple FailOver. Gonna play a little bit more and report if theres anything important.

Notes:

- The just created VM is showing up instantly when browsing the datastore through the other host

- Free disk space of the volume isn't updated on the host when creating a VM through the other host on the shared volume

- Paths to mounted ISOs are broken on the other host (of course)

Greetings

Björn

(1) http://professionalvmware.com/wp-content/uploads/2009/01/image1.png (Same message - not my screenshot)

[Bjoern_Gies]

Hi again,

I just tried what's happening when i suddenly turn off Host#1 where the VM on the shared volume is running. To my surprise I was able to start the VM on Host#2 without any problem - except the msg.uuid.moved message. Didn't expect the whole "thing" to be that robust.

Of course there was/is luck the filesystem inside the VM isn't wasted.

Greetings

Björn