On Tue, Aug 06, 2002 at 10:06:37AM -0500, Karl Bongers wrote:
> > UDP has a lot less overhead and is much faster over a clean/fast network.
> 
> It just seems like they'd have to "re-invent" TCP in the implementation.
> And if you have to re-invent it, then you would have just as much or more
> overhead as TCP.
> 
> I can understand UDP for short blips of information, like SNMP,
> but file transfer(you would think) would be the perfect application for TCP.
> Obviously its not that simple, and considering there is an option to do
> NFS on TCP, I'll bet it's a bit contentious a subject as well.

1. NFS is not "file transfer", it is "pieces of file transfer".
   Probably the most used RPCs are read(handle, offset, size) and 
   write(handle, offset, size).

2. NFS is a stateless protocol. It is designed such as the server can
   crash or be rebooted and after the rebooot the clients will just
   resume operations as nothing happened.
   TCP is statefull. This is an "impedance mismatch" with the semantics 
   of the upper level protocol that slowed its adoption as the
   transport. Linux still doesn't have a working NFS-over-TCP server.

florin

-- 

"If it's not broken, let's fix it till it is."

41A9 2BDE 8E11 F1C5 87A6  03EE 34B3 E075 3B90 DFE4
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://shadowknight.real-time.com/pipermail/tclug-list/attachments/20020806/a24e0fec/attachment.pgp