[NCLUG] NFS question

[Mike Loseke]

Thus spake Christopher J. Keist:
> 
> Thanks, I just saw that you have to have that compiled in.  Also says  
> still in experimental stages.

 I can attest to NFS v3 over TCP working quite well. I have 30+ clients
in a compute ranch which are accessing applications and data over NFS with
no problems. Attribute caching, as with all NFS clients, can rear it's
ugly head during certain race conditions, but it isn't specific to Linux.

 Of these machines I have a few which are exporting data to Linux and
Solaris clients. These are mounted v2/udp (explicitly by the client) for
reasons other than NFS performance. At one point there was a reverse lookup
taking place which caused a mount to take way too long (2, even 3 seconds!)
and I backed it out to v2/udp for my own sanity. I haven't re-tested this
at this time.

 As you stated, you just have to enable it all during a kernel build if
you're rolling one by hand.


> On Wednesday, March 26, 2003, at 02:01  PM, John L. Bass wrote:
> 
> > "Christopher J. Keist" <CJ.Keist at engr.colostate.edu> writes:
> >>      100003    2   udp   2049  nfs
> >>      100003    3   udp   2049  nfs
> >>
> >> Anyone know why it would not be showing tcp for nfs?  Is this  
> >> something
> >> to worry about?  All my other file servers, all Solaris, show both udp
> >> and tcp nfs.
> >
> > NFS defaults as a UDP protocol, a natural side effect of being a  
> > packet oriented
> > service rather than a data stream which is more suited for TCP.

-- 
   Mike Loseke    | If you thought Linux Advocates were obnoxious now,
 mike at verinet.com | wait 'til you see the new, improved Linux Amiga
                  | Advocate! -- Anonymous Coward on Slashdot