Tasked with setting up LTSP for a production environment, I'm interested to know if DFS is considered to be as reliable as NFS. I should mention connections will be fewer than 50, no streaming. Any feedback?

DFS is the shared storage solution for the PSSP server clusters from IBM. IBM is one of the most conservative companies in the world, and you can generally count on their stuff to last well.

I used an AIX v4 SP2 PSSP cluster between 1997 and 2007 at a US Steel mill (you know, back before the state and justice departments put the U.S. steel industry out of business. They are all now foreign owned.) and it was a mini mainframe replacement. Very solid.

That said, there are some 'gotcha' factors. Make sure you read the documentation, and note all cautions.

Disclaimer: I have NOT used the Linux implementations of DFS. Based upon experience in other areas, I would expect it to be just as good, but I cannot testify to that.

BTW: there is a Windows flavor of DFS. Samba support is very interesting, but not yet complete (mostly beta, in patches, and without DFS-R. So far!). Note that the Microsoft flavor of DFS is NOT really closely related to the original Unix/AIX DFS. As always, with Microsoft. Take care in reading online to note which DFS is in discussion.

Might be jumping in over my head but everything I'm finding on DFS suggests that it take's a handful of samba shares and puts them in one place (basically). You don't want to run a linux based system from samba shares.

That sounds like the MS implementation, and I would not trust that as far. The original DFS did not require or involve SAMBA in any way.

It makes sense to use some kind of DFS (or any cluster file system) for DATA storage that is in COMMON over nodes, not for BOOTING a node. You do not boot or run the node from that storage, you use that as SHARED storage.

It seems to me we have much newer and more modern distributed file systems for Linux now. DFS should be deployable, but if I had a project for it I would experiment with more recent tools first.