NFS Question

[Ken Ambrose]

On 29 Aug 2002, Kenneth E. Lussier wrote:

> 1) If applications are mounted via NFS, and someone runs the
> application, is the processor on the NFS server or the client system
> used? (I'm pretty sure it's the client, but I was told different today,
> so now I'm not sure..)

Client.  Though, of course, the server uses some CPU just in serving the
files, but only for NFSd -- _not_ the application.

> 2) What sort of problems are there with NFS on Linux? I have heard that
> there are file locking problems, but nothing really in depth. Anyone
> care to elaborate?

Some other OSen (eg. Solaris, HP/UX) whine about file locking.  There -is-
file locking in Linux's NFS, but, sometimes, not quite the way that the
other guys expect there to be.  From what I understand of the situation,
it's more a "you're not Solaris" problem than "you're a Linux box" problem.

Two other things: NFS under Linux (as Paul would be glad to point out)
ain't perfect -- maddog went on at a fair bit of length about this last
night.  It's a bit pokey, it's also a bit kludgey, and, last but not
least, not very well maintained (after all -- it's not sexy).  I've had
relatively few problems with NFS on a Linux server, though I'll admit that
last Friday my server _crashed_ because damn HP/UX decided to send two NFS
requests on port 50000-odd.  NFS requests are supposed to be 1024 and
below.  It was duly noted in syslog ("NFS request on insecure socket", or
somesuch), and then -whacko-: a nasty oops that made the system hang.
Ick.  My /etc/exports file now has "insecure" as part of the paramaters;
after all, anyone who assumes that port-based security is any kind of
security these days is merely deluding themselves.

$.02,

-Ken