Path: utzoo!attcan!uunet!cs.utexas.edu!sun-barr!newstop!exodus!terra.Eng.Sun.COM!brent
From: brent@terra.Eng.Sun.COM (Brent Callaghan)
Newsgroups: comp.protocols.nfs
Subject: Re: NFS over wide-area networks?
Message-ID: <2577@exodus.Eng.Sun.COM>
Date: 12 Nov 90 18:56:59 GMT
References: <6989@suns4.cel.co.uk>
Sender: news@exodus.Eng.Sun.COM
Lines: 30

In article <6989@suns4.cel.co.uk>, ajy@cel (Andrew Yeomans) writes:
> Is it possible to run NFS protocols over a wide-area network?
> Using UDP? Or TCP to guarantee order of packets?

Yup, you can do it with current UDP implementations but you
must be careful to set the rsize and wsize mount options
down to 1k to avoid fragmentation/reassembly problems.
You should also set the timeo option to a value that reflects
the increased round-trip-time particularly if there's a
satellite in the circuit somewhere.  Be aware that client
and/or server generates UDP checksums there's no way
to verify that the data received is correct.

I/O performance will be about as good as you would expect given
the bandwidth of the link and the transfer size.  Some non-IO
operations may be unexpectedly sluggish though - the NFS
protocol was designed to be used in local area networks with the
expectation of fast round trip times for request-response.  A
good example is a pathname lookup that translates into multiple
NFS lookup requests that must be executed synchronously.

IO performance of a implementation that uses TCP will be better 
than a UDP for WAN because it permits a larger transfer size
to be used without the penalty that UDP fragmentation/reassembly
suffers over an unreliable medium.  
--

Made in New Zealand -->  Brent Callaghan  @ Sun Microsystems
			 Email: brent@Eng.Sun.COM
			 phone: (415) 336 1051