What Do You Want From A Network File System?
We’ve been using OpenAFS as the School’s network file system (i.e. the thing that lets you access your DICE home directory and research group space on pretty much any machine from pretty much anywhere in the world) for the best part of 15 years. That’s a awfully long time in the fast moving world of computing and so one of the development projects I have on my plate at the moment is to look into whether OpenAFS is still the most appropriate fit to the School’s needs.
From my ivory tower, I can draw up a long list of filesystem features and capabilities which I think might be desirable but at the end of the day, what’s useful and what’s unnecessary can only be determined by you, the end user.
So please take a few moments to let me know what you think about the existing School network filesystem. Tell me about what you like about it, what you dislike about it and missing features you would like to see in a replacement. There’s no guarantee that any requests can be met but at least they can be taken into account. You can make your opinions heard by leaving a comment after this article or emailing me at
cms @ inf.ed.ac.uk
Remember, it’s the School’s network file system I’m interested in, your DICE home directory and research group space stored on the School’s file servers and accessed via a pathname beginning /afs/inf.ed.ac.uk/…. I’m not concerned at the moment with any centrally provided file space or data stored on self-managed machines.
I look forward to hearing from you!
Craig.
I know I’m a dinosaur, but I’m quite happy with AFS. It meets my needs. It took me time and effort to set it up on my (too) many machines, so I don’t want to invest that effort again unless there is a real gain in functionality that I would find useful. Because “15 years ..[is an].. awfully long time in the fast moving world of computing” is not a good enough reason in itself.
‘Age’ of the system is not the problem per se; however a fundamental problem is what seems to be diminishing active support for the OpenAFS project.
Gosh, is it 15 years now in which I’ve never found a use for the ability to “access [my] DICE home directory and research group space on pretty much any machine from pretty much anywhere in the world”? I remember trying to understand how this was supposed to work and why one would care, once, but really the only impact the NFS-OpenAFS switch had on me was that I no longer really understood the file access control capabilities, sigh.
I’m not sure which of the things I want are really properties of a file system per se – peraps a bit more guidance on what information you want and what the options are would be useful? I want my stuff stored in a way which is safe (won’t get lost or damaged) and secure (can’t be accessed by people I didn’t think could access it, including, I’d like to be confident I can tell who can access it). Beyond that I struggle to come up with requirements. I guess this says I don’t care whether we have a *network* file system at all, let alone which one?
well, that goes too far of course – it is handy to be able to access my home directory from lab machines as well as from my office!
I do *mostly* like AFS – from a user perspective, it solves a range of problems, and normally works well. I’ve successfully got it set up on a variety of self-managed machines, utilising the School’s kerberos infrastructure – BUT it seems to throttle me when doing file intensive operations, which is really annoying.
For example, using “npm” and doing an update or an install, involves a significant amount of file operations, and AFS grinds to a halt, taking hours and hours (instead of 10 minutes). I assume that I’m being throttled – I couldn’t find any other reason why this should happen, and I don’t know if there is a solution, but it really limits the kind of workloads that I can do.
[better late than never?]
The directory-level-only permissions model has at least one serious bad side-effect: no hard-linking of directories, or of files in different directories. I don’t run in to it often, but when I do it’s a real pain. It means AFS and conda don’t play nice with each other.
The frequency with which the so-called “AFS getcwd” bug forces me to reboot has slowly gone down, but it still happens (last month most recently), and is a pain. See e.g. https://rt4.inf.ed.ac.uk/SelfService/Display.html?id=86776
IIUC, this is also an unintended consequence of AFS fighting with itself over mount paths.
I also have found no use for the “mount from anywhere” property, and stopped configuring my Windows laptops to try to mount my home and group AFS directories many years ago
I also find the granularity of the access control unhelpful: I got burned only last week because rliw does _not_ allow people to rename, because renaming amounts to deletion (Always, or only sometimes? Quick! Can you answer without looking it up?)
So, what’s left that I want?
* Seamless single-login availability within the School subnet;
* Uniform access control mechanism across local disk and network mounts;
* ‘Yesterday’ functionality.
Thanks