Our recent Webcast: NFSv4.1 – Plan for a Smooth Migration was very well received and well attended. We thank everyone who was able to make the live event. For those of you who couldn’t make it, it’s now available on demand. Check it out here.
There wasn’t enough time to respond to all of the questions during the Webcast, so we have consolidated answers to all of them in this blog post from the presentation team. Feel free to comment and provide your input.
Q. Will NFS 4.2 be any easier to migrate to than 4.1? Would it be worth waiting for?
A. NFSv4.2 is a set of additional functionality that will be easy to take advantage of – if you’re on NFSv4.1. The first move is to NFSv4.1, as it offers a wealth of features over and above NFSv3. Waiting for NFSv4.2 features wouldn’t be advisable; it’s unlikely to be ratified until the end of 2012, and enterprise server solutions and the required downstream client distributions will be a lot further out than that.
Q. Since NFS 4.1 is out, what is the uptake in the industry?
A. There aren’t any global figures, since not all suppliers collect detailed information about protocol usage, and of those that do, many can’t differentiate between NFS versions. Anecdotally, it’s slow. That’s because NFSv4.1 servers (particularly for file-layout) have only been available for less than a year, and the needed Linux client support has only recently made it through to the enterprise distributions.. NFSv4 (as opposed to 4.1) is more widely used; but the only figures I have are anecdotal, and would be misleading.
Q. Are there any network architecture design considerations that need to be taken before implementing NFSv4.1?
A. No. In fact, (if you’re not using pNFS) NFSv4.1 should get you more “bang for your buck” as there’s a reduction in network traffic compared with NFSv3. pNFS requires a different architecture; your storage vendor should be able to assist in the planning.
Q. Clustered servers – you mentioned that vendors had to provide a special server for this… are these enhancements going to be ported into the general linux nfs server stream?
A. I’m not sure to what this refers; perhaps the MDS (metadata server)? Although this server is often shown as a separate box in diagrams for simplicity, that’s not how it is normally implemented. The MDS is normally part of the cluster running on one or more fo the data servers.
Q. If you recommend AD for kerberos, do all of the NFS clients need to be joined to the same AD domain as well? Or only the servers?
A. Any time a client in one domain (or realm) attempts to access a server, the server must be in the same realm as the client, or if it’s in another realm, there must be cross realm trust so that the principal (the client) can be correctly authenticated.
Q. Can you talk about any difficulties in using Active Directory with NFS? Are there changes needed on AD?
A. No changes are needed to AD. It’s relatively straightforward security administration, and storage vendors should be able to provide you with implementation checklists.
Q. What is the impact on clustering and failover by introducing statefulness?
A. Significant! And much better. Recovery is much improved, as the server and client after a failure can attempt to agree on what locks were held, what files were open, what data had been written and so on. It’s a big improvement on NFSv3.
Q. Will it be possible to mount root file systems from NFSV4? Like boot from the SAN that we already have in FC or iSCSI?
A. Yes, that doesn’t change.
Q. Can you explain the reasons why home dir and hpc would benefit with v4.1?
A. Home directories are an easy win; no application (well, at least that you care about) and easily migrated. The same is often true of HPC. For example where the data is transient – served from a store to local disk, computed and crunched, and then sent back to the store – the store could be migrated to NFSv4 and the app later; or the app first and the store later.