Should I have more than one shared volume (LUN) on my storage device in a cluster?

[johnnyIsVirtual]

This is similar to a recent question, but I think it is different.

If I have two volumes on the same SAN and two cluster nodes mounting one of the volumes, and I have all the VM guests on the same volume, and that volume dies, not the hosts, the underlying storage volume on my SAN (not the SAN, just the volume/iSCSI LUN), am I correct I lose all my VMs?

If that is correct, is there any reason I need more than one SAN volume for my VMs, instead of one giant disk volume, iSCSI LUN, mounted by the cluster nodes?  I can't think of a reason I need more than one volume as opposed to one big volume managed by the VMware clustering software.

[daphnissov]

If I have two volumes on the same SAN and two cluster nodes mounting one of the volumes, and I have all the VM guests on the same volume, and that volume dies, not the hosts, the underlying storage volume on my SAN (not the SAN, just the volume/iSCSI LUN), am I correct I lose all my VMs?

Pretty much.

If that is correct, is there any reason I need more than one SAN volume for my VMs, instead of one giant disk volume, iSCSI LUN, mounted by the cluster nodes?  I can't think of a reason I need more than one volume as opposed to one big volume managed by the VMware clustering software.

There are several. For performance reasons, having multiple extents means separate queues. For data services features, one LUN could be replicated and encrypted while the other is not. Each are backed by the same array. For capacity reasons, having multiple extents means more effective used capacity without triggering alarms.

[johnnyIsVirtual]

Not questioning you here, only trying to understand.

I don't understand why if I have 5TB why I'd need VM1 to be 2.5 and the other, say VM2, to be 2.5TB volumes.  Should there be copies of the VMs on the other volume, mirrored?  I'm not following, and I was hoping you could tell me why having one large volume on the SAN can cause problems.  Supposedly, the SAN can do all the load balancing across the controllers.  If I have one volume that has all the VMs and one volume that has, well, nothing on it, I'm not sure I am doing things right.

Also, I couldn't see where VMFS does any kind of "striping" across volumes for redundancy (something like ASM on Oracle).

[daphnissov]

I don't understand why if I have 5TB why I'd need VM1 to be 2.5 and the other, say VM2, to be 2.5TB volumes.

No, I'm not advocating for distributing a single VM across multiple datastores. That is generally not a recommended practice. If you have a 5 TB VM, you should try and put that on a single datastore unless you have specific requirements. This obviously means you need a single datastore of more than that capacity.

why having one large volume on the SAN can cause problems.

It's not that necessarily it would cause problems, just that there are reasons for not doing this. A fourth reason is blast radius. If someone screws up that one LUN and it has everything on it, you're completely hosed. Even if you have backups, now you have to restore absolutely everything versus, say, 20%.

Supposedly, the SAN can do all the load balancing across the controllers.

Not talking about array-level balancing when it comes to queueing, talking about ESXi and VMFS queues which are per LUN.

[johnnyIsVirtual]

Sorry I wasn't clear.  VM1 was the name of a volume, not an actual VM guest.  The 5TB was for a volume that holds all the VMs.

But if someone does hose up a LUN, then I will lose the VMs on that LUN.  I guess you are saying the other LUN might have other guests that were redundant servers for the whole system, yet another layer of abstraction.  Maybe the e-commerce system has multiple httpd servers.

Other than that I can only guess the other LUN should have a copy of what is on the first LUN.  A lot of what I have are just one off VMs because that's all I needed.

[daphnissov]

I guess you are saying the other LUN might have other guests that were redundant servers for the whole system, yet another layer of abstraction.  Maybe the e-commerce system has multiple httpd servers.

Either that or just separate VMs totally. Just because my Sharepoint server goes down doesn't mean my Jenkis server has to.

Other than that I can only guess the other LUN should have a copy of what is on the first LUN.

No, it shouldn't. Not only is that wasteful but it does nothing to help in cases of logical corruption. This is what backups are designed to do, and those backups should be stored on a totally separate system.

[johnnyIsVirtual]

Then it seems outside of queuing problems or some other need, one volume would be fine.  If I don'[t have duplicates on the other volume, and I don't have overriding performance or architecture needs, I don't seem to need that volume just taking up space.

[IRIX201110141]

There are many reasions for having multible Datastores instead a single one

- Queue

- SCSI Reservations

- Dont put all eggs into one basket

- Using "new" FT creates a copy of all vDisks... and you should select a (different) Datastore for that

- Maybe another snapshot schedule on SAN side because of different needs

- Depends on the SAN.... a LUN is often assigned to a storage controller. If you only have one LUN all the load handled only by one controller.... the other idle around

Regards

Joerg