How to plan SAN Storage?

Question

I'm pretty new to the whole SAN thing and am trying to wrap my head around some fundamentals. We are trying to build up a Hyper-V Cluster with two Servers and some external storage media like an HPE MSA 2050 (not set in stone), but I have some very basic (and maybe stupid) questions:

1. Can i simply connect the MSA directly to the two servers without having to buy a dedicated SAN-Switch (preferably over FC)
2. How many storage do I need? Let's assume I want to have 40TB available (disregarding parity). Do I only need one set of drives in the SAN and both servers can read and write the same blocks, or do I need everything "doubled" so each server has it's own set of drives? I have read a bit about CSV (cluster shared volumes) but am not sure if I understood it correctly.
3. The HPE SAN I named claims to have two controllers and with it also having two PSUs, does that mean it is redundant enough in itself that you wouldn't need a second one (assuming regular failouts and not WW3 bombing the entire serverroom)

I hope you can help me with this, thanks in advance.

Answer 1

As already mentioned, for a 2-node cluster, SAN is overkill. Plus, despite redundant controllers and PSUs, it's still a single point of failure. Go for local drives plus some VSAN software instead. For example, take a look at Starwind VSAN: https://www.starwindsoftware.com/starwind-virtual-san. It's been fast and stable for me on several 2-node clusters. There is of course Microsoft-native Azure Stack HCI (it's still S2D undercover): https://azure.microsoft.com/en-us/products/azure-stack/hci/#overview. However, when I've tried it on 2-node setups multiple times, it simply couldn't withstand a Windows update without losing storage. I had no such issues and it worked well on 4+ nodes to be fair though.

As to your questions:

1. Yes, you can use FC HBA cards directly in servers.
2. So you need 40TB. Count it as you would with a simple RAID. No need to double. You can present separate LUNs to each host but if you want both hosts writing simultaneously to a shared LUN, you would need a clustered filesystem (CSV) and correspondingly a Failover Cluster: https://docs.microsoft.com/en-us/windows-server/failover-clustering/failover-clustering-overview. Otherwise, with a common filesystem, there is no file-locking and one node is not aware that the other one is writing at the same time so it will just lead to data corruption.
3. There are other things that can happen to a storage box besides a controller failure. I would say it's as redundant as a single storage box can be.

Answer 2

Can i simply connect the MSA directly to the two servers without having to buy a dedicated SAN-Switch (preferably over FC)

Yes you can, obviously if you use FC then you'll need FC HBAs in both servers, and you'll need to configure the ports on the MSA to be in 'direct connect' mode rather than assuming they're connecting via FC switches, but this method works perfectly, I've done it myself in the past using 2040's. If you're using iSCSI via Ethernet then you can simply connect the ports to your switch and get to it that way.

How many storage do I need?

This is a bigger question I'm afraid and the answer is likely to vary a lot based on your detailed requirements.

The easy bit is yes it's very easy to get 40TB usable - of any type of disk too - they can even take 14TB HDDs! Just make sure if you're using HDDs instead of SSDs that you DON'T use RAID 5, it's considered dangerous these days - stick to R1/10 or R6/60, they'll look after your data :)

The HPE SAN I named claims to have two controllers...

Probably yes, if you can then try to drive the two separate PSUs off different UPSs, on the same phase of course, that'd help, but yes you can survive the failure of a single PSU and even a whole controller so long as you've setup multipathing (one or more paths per controller).

Anyway good luck with this, come back if you have more detailed questions in the future ok.

Answer 3

1. Can i simply connect the MSA directly to the two servers without having to buy a dedicated SAN-Switch (preferably over FC)

Yes. As long as you've got ports on the appliance you don't necessarily require a switch.

2. How many storage do I need? Let's assume I want to have 40TB available (disregarding parity). Do I only need one set of drives in the SAN and both servers can read and write the same blocks,

Technically, that would be possible. However, simultaneous access to a LUN requires a clustering file system, coordinating concurrent access from the client hosts. If you run simultaneous access to a normal NTFS, ext4, ... volume, data will get overwritten and destroyed in no time.

Block storage is shared as LUN objects. Each LUN represents a logical disk to the client host. In your scenario, you could create a single disk array across all physical disks, or two arrays to utilize dual controllers, then create multiple LUNs on those arrays and finally map each LUN to one of the hosts. Make sure you don't use RAID 5 for disks larger than perhaps 300 GB.

or do I need everything "doubled" so each server has it's own set of drives?

Yes, basically.

I have read a bit about CSV (cluster shared volumes) but am not sure if I understood it correctly.

That is a cluster file system that allows concurrent access from multiple hosts to a single LUN. Depending on the workloads, the appliance features and the storage protocol (FC or iSCSI), a cluster file system may perform rather poorly when compared to dedicated access.

I wouldn't use a cluster file system for general file shares but it's ideal for storing VM disks (since each disk is locked just once, when the VM boots) and has the benefit that you don't need to move any data when migrating VMs between hosts.

3. The HPE SAN I named claims to have two controllers and with it also having two PSUs, does that mean it is redundant enough in itself that you wouldn't need a second one (assuming regular failouts and not WW3 bombing the entire serverroom)

Yes and no. Dual controllers and dual PSUs enable operation without any single point of failure. Additionally, you'd need dual UPSs running on different power phases, and redundant cabling to your hosts for fully redundant operation. "Redundant enough" depends on your requirements.

A basic setup would look like this:

• Each MSA PSU is connected to a different UPS.
• Each UPS runs on a different power phase.
• You create two arrays on the MSA, each one mapped to a preferred controller.
• You create one LUN on each array.
• Each host has got two storage ports, each connects to a different controller.
• If you use storage switches, each host-to-storage link runs over a different switch path.
• Map each LUN to one of the hosts.
• If required (without ALUA), configure each host so that its primary storage link is the one to the preferred controller for its LUN.

Answer 4

You should not consider using a SAN for something like this to start. Use locally attached storage in a windows high availability cluster. Faster with a proper network and one less element.