Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!purdue!haven!adm!smoke!gwyn
From: gwyn@smoke.BRL.MIL (Doug Gwyn )
Newsgroups: comp.unix.questions
Subject: Re: Several questions
Message-ID: <9919@smoke.BRL.MIL>
Date: 24 Mar 89 15:37:56 GMT
References: <1316@blake.acs.washington.edu>
Reply-To: gwyn@brl.arpa (Doug Gwyn (VLD/VMB) <gwyn>)
Organization: Ballistic Research Lab (BRL), APG, MD.
Lines: 16

In article <1316@blake.acs.washington.edu> icsu6000@caesar.cs.montana.edu (Jaye Mathisen) writes:
>1)  NFS is supposed to be a "stateless" protocol.  What good is a stateless
>protocol that requires  a whole schmeer of stateful servers to  provide
>"most" of the semantics of the Unix file system?

By not having to propagate revised state information around to all users
of a file whenever any one of them makes a change, supposedly efficiency
is improved.  Also, core NFS deliberately lacks some UNIX file system
semantics to make it compatible with non-UNIX file systems, MS-DOS in
particular.

In a UNIX environment, I agree with your implication that the RFS
(stateful) approach is generally better.  I especially like RFS's
preservation of device driver semantics across multiple hops.  I never
did hear how AT&T was going to solve the problem of ioctl data format
incompatibility among heterogeneous systems, though.