Current-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: Coda vs NFS



Miles Nordin <carton%Ivy.NET@localhost> writes:

>>>>>> "pw" == Patrick Welche <prlw1%cam.ac.uk@localhost> writes:
>
>     pw> Idle question: is there anything coda can do in theory which
>     pw> nfs4 can't?
>
> I have never used AFS/Coda but aren't they very different in
> performance from NFS?  Everyone focuses on the nice-soudning aspects
> of cacheing, but for example, when you open the file, don't AFS/Coda
> copy the whole thing into the ``cache'' filesystem which is something
> like FFS, and then map all file access into that FFS?  Then copy back
> changes when (and only when!) you close the file?  or am I wrong?

Yes, basically you have it right.  With AFS, I think it might do
partial-file caching for reads, but I'm not sure.

> If that's the case, using AFS or Coda would work well for reading
> email or maintaining a tree of .c files, but would ~ not work at all
> for an .avi collection, a large database, or a VMware/VirtualBox disk
> image store where the unpathological case requires not transferring
> the entire file over the network.  like, AFS/Coda would just crack and
> break under this workload, would it not?  It is not really a NAS at
> all, just a backup/collaboration tool.

It is a distributed filesystem, which is different from NAS.  It's true
that really large files are the wrong workload for it.

> It's hard to get a straight answer so maybe I've got it all wrong.

I'm surprised; coda people are usually pretty straightforward about this
sort of thing.

> nfs4, fwiw, is not really different from nfsv3.  It's supposed to
> perform better, but often has lots of bugs some of which make it
> perform worse, or not keep working across a server reboot like it's
> supposed to.  It's supposed to come with this new mess of ACL's, but
> they are inspired by Windows ACL's so it's like smb.conf and no one
> really udnerstands them, often break things because solaris likes to
> ``fabricate'' ACL's and clients support them only partially.  There is
> talk about support for server replication and client failover, but it
> does not exist yet, not even in the solaris client---only the ``mirror
> mounts'' thing is working which is just an excuse to avoid using the
> automounter.  Many years after its release its support is pretty
> rare---even under solaris I cannot get it to mount '/' over
> nfsv4---keeps coming up as v3 but then you can have it mount /usr over
> top of it as nfsv4,...so...job is not really finished...but no one is
> in much of a hurry so no worries.  IMHO nfsv4 is more like the
> inevitable future than the revolution AFS and Coda meant to be.
>
> pNFS is something different and actually claims new features: split
> data/metadata for storing big files like video, database, disk image
> in clusters.  so the true relevant nfsv4 changes were probably laying
> the path to pNFS.  but i don't think pNFS exists yet.

Thanks for explaining.  In the end, I think they are different tools for
different jobs.

Attachment: pgpF2taSh3q8n.pgp
Description: PGP signature



Home | Main Index | Thread Index | Old Index