So, quick question to the Windows networking and storage experts. I need a way to have 2 servers offer the same UNC path to network clients. So that if one of them goes down, the other one takes over. To achieve the "cluster" connectivity itself to point UNC to either server I simply use DNS Round Robin technique and programmed a small tasktray agent that runs ipconfig /flushdns on the clients if one of the two servers cannot be pinged, as a sign that it's time to change. This works fine as the domain updates immediately to correct server IP that's still up - back and forwards. Tested OK. The UNC then points to the other server that's up. What I haven't figured out, is how to synchronize/replicate the shared folders between them properly. I want each server to be a full backup of the other, ready to take over automatically if the other goes down. And I only have 2 physical servers to set it up with. Currently there's a VM with the shared folder using Hyper-V replication from one to the other using workgroup SSL, but this requires manual failover once one of them goes down. I need automatic cluster-like/load-balancing behavior so the users still have service without waiting for tech support if a server goes down. What I've been reading about windows networking, it seems nothing will offer exactly what I'm thinking of without going shared storage (SAN/NAS). DSF-R (distributed file system replication) will seemingly crash if a member goes down, and it requires AD as well which I have no experience with but will set up if it helps. I simply need a replicated shared folder to be available on arbitrary server that's still up, if one goes down. And if it comes up again, I need them to synchronize/replicate eachother before the share becomes available again. < This is the main part I haven't figured out yet. I've been thinking about pausing lanmanserver while it syncs with robocopy or some other program, but figuring out exactly when to do it and how to trigger it is messy. Is the only way a shared RAID/SAN/NAS storage server/unit? Or is there really a way to keep 2 servers fully replicate a simple shared folder where clients always work with the newest files, regardless which server they connect to. Using a SAN/NAS shared storage just makes that storage another single point of failure again, so trying to figure something out using just 2 physical, identical servers. Or, could I set up both servers as a storage cluster for eachother somehow? They're running 2012 R2. The strict limitation I have is the 2 physical servers with 2012 R2 on each. I just wanna make the most out of them. As automatic as possible. Maybe it's simply impossible. As if host1 goes down, host2 takes over for a while. Host1 comes back, but if Host2 then is down, Host1 only has old files as it can't sync back updates. Halting both servers. This is an unavoidable potential MURPHY problem if I'm going to do it this way, right? Best to go NAS/SAN for storage regardless perhaps? Thanks.
The entire post is about how shared storage is needed, not that I need it. That was fairly obvious. I get that clustering the shared storage is actually a rather new concept for most people (entirely different from just VM clustering which requires that exact shared storage, with people mostly ignoring that the shared storage is a SPOF, which is what I'm focusing on). I've already looked closely at StarWind VSAN which you're linking to and it costs 2500 EUR to run hyper-converged, not free anymore unless 2 more servers are added, which would bring it easily to 5000 eur - holy *. But yeah it would have done what I wanted, in theory. However ALSO in my post is the wish to do it with only 2 physical servers, the same which function as VM cluster nodes. Their price is way above reasonable for what I want to do. Not sure how well it would work but also looking at StableBit. Their DrivePool solution is battle tested and very solid. Using it together with their CloudDrive solution may do what I want.
Job not done with a iSCSI/SAN storage array since it becomes a SPOF, which is a part of the whole point to avoid. Pointless having a VM cluster if the array itself goes down because of bad controller or any other non-disk hardware failure - it's the dumbest HA structure ever to rely on as a simple SAN failure will crash the entire cluster. Having shared SAN/iSCSI/block-based replicated storage across the nodes that functions as a cluster already, would be ideal as no more hardware is needed, and the nodes would be fully independant. I'm aware Server 2016 has more native support for what I'm looking for, but I'm stuck with 2012 R2. Might have to look at upgrading.
You're just repeating yourself with your SPOF - run of the mill - setup. Good enough for enterprise you say, that was RAID 5 and 6 just a few years ago as well which is still being used ignorantly today. You don't want to do that anymore either on new setups. It's okay, I was looking for actual input for what I asked for anyway, not what you personally think is "good enough". I've accepted your opinion by replying to it. Thank you for stepping off.
Is this not simply DFS. I believe the Distributed file system does handle it. I've used DFS for domains and multiple shares. DFS now also handles the replication which I have also used. I've just not used 2 servers on the same share, but I did a quick DFS search and it did seem to work. Since this is a windows server forum, I am guessing you are on a windows server, so it should be worth at least a look.
Yeah I'll be trying out DFS-R more to see what I can expect from it over time, maybe. I'm using 2012 R2, with hopefully a lot of improvements to it from older versions. Nags on me a little that it requires AD though, shouldn't have to. Same with clustering. I think they removed or lessened that requirement in 2016. But upgrading server OS is not possible ATM.