Best solution for a hyper-v stretched cluster

[Bohgdu]

Hi!

I'm currently needing to up the resilience on one of our on-prem Hyper-v clusters and I'm in need on someone else perspective on how to best solve my problem.

We are currently running 4 nodes and one dell ME5024 iSCSI as storage.
Now comes my predicament, i would like too move 2 of the nodes to another building and run our fail-over cluster in a stretched configuration, my plan is to also add a witness in a third location. I have been given a second ME5024 from another team, so I'm able to have storage at both sites. However looking at the ME5024 it does not support high availability, so I'm unable to present the storage as on unit to my cluster. I've been looking at Microsoft solutions for this using storage replicationhttps://learn.microsoft.com/en-us/windo ... ed-storage. Unfortunately it does not appear to act in the way i would like in terms of VMs being able to fail-over without interruption between the two sites.

So I'm currently looking for options on how to solve my issue. Is there anyway for me to use Starwind software to "virtulize" the two dell Sans and therefor get active-active replication or would a better option just be to butcher the ME5024 and take the drives and put them into my servers and run something like S2D or Starwind vSAN?

[yaroslav (staff)]

Hi,

Welcome to StarWind Forum!
In a nutshell what you explain is doable, yet I would like to dig into more technical details, like host and link spec, node arrangement, etc.
StarWind VSAN provides highly available shared storage. It does not move the VMs or virtualize the host's compute resources. From the storage perspective, failover happens instantly, as VSAN provides shared active-active replicated storage.
When it comes to the VMs, their move is handled by the Microsoft Failover Cluster. None of the shared storage solutions can override its logic, it can just make sure the storage is there when it is time to fail over.
See more about VM availability https://techcommunity.microsoft.com/t5/ ... a-p/372027
I would strongly recommend a demo and trialing the solution. Fill in this form https://www.starwindsoftware.com/v17-request-live-demo

[Bohgdu]

Thanks for the quick response!

Our four nodes are Dell r750 each with three 2x25Gbe Broadcom nics, so a total of six 25Gbe uplinks per server. Today we are using 2 for inbound traffic and 4 for connecting to the SAN, Hyper-v fail-over and some other management. We are planing to ad redundant 100Gbe links between the old and the new site to be able to keep up with the write performance of our disks.
Today as everything is in one site with the one ME5024 providing storage the cluster can live fail-over VMs between the host without causing any interruption for the services and that's what i wanted to keep as we moved half the cluster to another site. I had hoped when i was given the second ME5024 was that i would be able to run both SANs in and active-active role, therefor keeping the live fail-over feature. However as the two sans don't have HA, you need to buy higher teir SANs from Dell to get that feature, I'm having to find another solution.

I'm somewhat familiar with Starwind vSAN as I run it in a two node hyper-v cluster and have been for some time and I'm a really big fan as it's been running solid for more than a year and that's what brought me here. after doing my post I did some reading about your SAN and NAS software(I'm now also seeing it's a legacy product), but as i read through it it was more aimed at turning you standard server into a SAN.
I had hoped to be able to do some software magic and get some use out of the two ME5024, but i guess it's a better option to scrap the two ME5024 and put the disks directly into the nodes and create a vSAN solution.

[yaroslav (staff)]

Hi,

Thanks for your update.
I am afraid, there is slight confusion here. Do you mean live migration by the failover? If so, the cluster definitely will be able to do it with StarWind VSAN.
I was talking about the situation where the server fails and VMs fail with it and need some time to migrate.
SAN and NAS is a part of StarWind CVM. It did not go anywhere, but just got reshaped

but as i read through it it was more aimed at turning you standard server into a SAN.
I had hoped to be able to do some software magic and get some use out of the two ME5024,

That's right but it does not give HA capability.
Standard, non-replicated devices can help turning a server into a SAN. Yet, I think your setup needs High availability.

A stretched cluster will need either direct links or redundant switches on site. If you cannot provide redundant switches/, you can spin up the Witness node in the third location.
I just would really want to be sure the setup meets https://www.starwindsoftware.com/system-requirements and https://www.starwindsoftware.com/best-p ... practices/
See more on possible configurations heartbeat (https://www.starwindsoftware.com/help/H ... ategy.html), node majority (https://www.starwindsoftware.com/help/N ... ategy.html). For both cases the latency has to be less than 5 ms.

[deanthomas]

You could use StarWind to virtualize the two Dell SANs for active-active replication, minimizing VM failover disruption. Alternatively, consolidating the drives into your servers and using S2D or StarWind vSAN might offer better integration and performance, depending on your comfort level with software-defined storage.

[Bohgdu]

You could use StarWind to virtualize the two Dell SANs for active-active replication, minimizing VM failover disruption.

That was one of my hopes when I first started looking for a solution, but I don't know if that is even possible. The only way i see that happening is if I'm able to install another firmware/OS on the SANs or maybe try and turn them into NAS-boxes and put Starwind on that.
But my plan for now is to put the drives into the servers and go with Starwind vSAN. My problem now is trying to find disk caddys so that i can move the drives from the SANs and into the servers, Dell wont sell me just the caddys.

[Bohgdu]

You could use StarWind to virtualize the two Dell SANs for active-active replication, minimizing VM failover disruption.

That was one of my hopes when I first started looking for a solution, but I don't know if that is even possible. The only way i see that happening is if I'm able to install another firmware/OS on the SANs or maybe try and turn them into NAS-boxes and put Starwind on that.
But my plan for now is to put the drives into the servers and go with Starwind vSAN. My problem now is trying to find disk caddys so that i can move the drives from the SANs and into the servers, Dell wont sell me just the caddys.

Wait, I'm not thinking straight. I get what you mean by putting vSAN on the SANs. I'll direct-attach the SANs to the 2 nodes on each site and put Starwind vSAN on that. There will be some waste of performance as the SANs will have 2 copys of the data and have two write it twice but that might work. I'll have to do some testing and see what the performance is like.

[yaroslav (staff)]

Greetings,

Please keep me posted.
Please also consider trialing the product (i.e., use the trial key).

[Bohgdu]

After finally getting time for some testing , I was able to try out the vSAN solution and unfortunately I was not able to get the product to work as I had hoped.
As I was only able to have 3 nodes running synchronously and not 4, I was not able to achieve a loss of 3 nodes and still keep the vms running.
I will instead be looking at some powerstore sans from Dell as they’re able to run I in HA.

Thank you all for answering my questions.

[yaroslav (staff)]

You are welcome. Sad to see you go
StarWind VSAN does not support mirroring between 4 nodes at this moment. What you can do is something like GRID https://www.starwindsoftware.com/resour ... hitecture/.
If you decide to give the solution a try, reach out to us at https://www.starwindsoftware.com/v17-request-live-demo

[justinb]

Using Microsoft’s Storage Replica can work, but as you noted, it may not provide the seamless failover you desire. It’s typically used for disaster recovery rather than active-active configurations. Adding a witness in a third location is a good strategy to maintain quorum in a stretched cluster.

[yaroslav (staff)]

Quorum is only for clusters with even numbers of nodes. In such a large cluster file share cluster witness could help.