homelab

482 readers

3 users here now

homelab

founded 2 years ago

MODERATORS

baascus

Options for SMB shares without a single point of failure. (self.homelab)

submitted 4 weeks ago by zhill29 to c/homelab

6 comments fedilink hide all child comments

Why I'm dabbling in this is a very long story, but lets assume I only have 2 servers at my disposal and want Windows Server VMs providing SMB shares without a single point of failure. Proxmox and HyperV are my options for hypervisors.

Ceph is out for a few reasons, most notably only having 2 servers, 1Gb networking and the Windows Server VM would still be a single point of failure. I've been reading up on Windows Storage Spaces and if I'm understanding correctly it seems I could cluster 2 physical or virtual servers, replicate the data between the 2 and present the SMB share as the cluster name/IP rather than individual servers.

Before I spend too much time setting up Windows clusters, are there any other options I should be looking into?

top 6 comments

sorted by: hot top controversial new old

[–] [email protected] 4 points 4 weeks ago (1 children)

What kind of failure are you looking to address? Data resiliency, availability, automatic failover, etc?

[–] zhill29 1 points 4 weeks ago

Automatic failover, basically should a VM lock up in a way that monitoring/HA failover isn't triggered can another VM be picking up the slack.

[–] [email protected] 1 points 4 weeks ago (1 children)

https://docs.gluster.org/en/main/Administrator-Guide/GlusterFS-Introduction/

I've heard good things about glusterfs.
Never used it tho, I just go with ceph

[–] zhill29 1 points 4 weeks ago

Looks like Gluster also requires 3 servers.

[–] slazer2au 1 points 4 weeks ago (1 children)

If you are keen on Windows does DFS NameSpaces fit the bill?

[–] zhill29 2 points 4 weeks ago* (last edited 3 weeks ago)

My understanding is that allows one server to present storage from multiple 'back end' servers, thus still being a single point of failure, right?

Maybe, it could be separate shares on separate servers presented as a single 'host' by a Windows cluster, this would be more storage efficient than replication and the only single point of failure would be any given back end server that would only affect 1-2 shared folders rather than all of them, which might be acceptable. Or I could be way the hell off with my understanding of DFS....

Edit: Did a bit more research, it seems DFS does do a redundant namespace that can handle failover. That might actually be exactly what I need. Thanks!