[rescue] Re: OT Linux

[Dave McGuire]

On December 25, dave at cca.org wrote:
> >  NFS defaults to UDP, and most NFS implementations *only* do UDP.
> >TCP can be better in some situations because it can take advantage of
> >stuff like congestion control...but UDP NFS mounts have the advantage
> >of being stateless, so you don't need to remount filesystems when the
> >server reboots.
> 
> Really? I thought it was only an issue of packet level reliability.
> 
> Perhaps that's why I've never particularly grokked NFS client behaviour
> during server reboots. :-)

  Well, think about it...NFS transactions are stateless, so NFS
doesn't care if the server goes away and comes back, since UDP is
stateless as well.  But TCP is *not* stateless; if you have a TCP
connection between two machines and one end goes down, the connection
goes down...and when the machine comes back up, the connection state
is lost.

  Now, if something on the client reestablishes that TCP NFS mount as
soon as the server comes back up, say, by polling it with ICMP or
something, then things will be fine.

      -Dave

-- 
Dave McGuire
St. Petersburg, FL         "Less talk.  More synthohol." --Lt. Worf