How should SAN shared storage be engineered between VMware ESXi cluster nodes?

[johnnyIsVirtual]

If I have two nodes in my cluster, and I have two volumes on a single SAN, both nodes mount both volumes and format them at the block-level with vmfs (via iSCSI), the cluster allows same block-level management so there is no data corruption.  Is that correct (that's shared storage in a cluster, is it not)?

My real question is if I have two SANs, should there be a volume on each SAN so that if one dies, VMware can migrate to the working SAN?  In other words, say I have shared storage in my cluster with two nodes, if both iSCSI volumes are in the same SAN, then that is a single point of failure.  Should my topology be one volume on one SAN and one volume on the other SAN.  Both nodes can still mount both volumes in the cluster just from two different SANs.

I do not have vSAN, and it is an older version 4.1 implementation, but the principles are the same, I believe.

Do I misunderstand this?

Thanks.

[a_p_]

VMFS is a real shared storage file system , i.e. it can (which is the usual case) be accessed by multiple ESXi hosts simultaneously.

Whether you use a single, or multiple storage systems, and how you replicate data depends on your needs and requirements. A single storage system may indeed be a single point of failure. However, most storage systems come with redundant controllers, so that one can fail without interrupting storage access. If you want to have a fully redundant storage system, you will also need to make the storage area network (switches) redundant.

vSAN on the other hand is different, because it virtualizes the hosts' local storage

André

[johnnyIsVirtual]

What do you think about the two SAN part?  Shouldn't I have one iSCSI mount on one SAN and one iSCSI mount on the other SAN and both/all nodes in the cluster mount both of those?  That way if one SAN dies, I can still migrate to the other SAN.

Granted the SAN totally dying is somewhat unlikely, but to me it did not seem correct to have all my datastores on a single SAN (assuming I have more than one SAN available.)

[a_p_]

As mentioned before, it depends on your requirements.

That way if one SAN dies, I can still migrate to the other SAN.

Well, you can't ride a dead horse. If  one storage dies, you woun't be able to access it anymore to migrate the data.

Having a second storage system is definitely a good thing, but in this case you also need to replicate the LUNs, or VMs between both of them (synchronously, or asynchronously), so that in case one of the storages fails, you can run/start the VMs from the surviving storage.

André

[johnnyIsVirtual]

I think I understand.  Most of the redundancy in VMware and migration is about hosts failing and automigration, not storage failing like on a SAN.  This would explain why that is all I see online in writing, videos, etc.  They always show nodes in the cluster going down, not the shared storage.