Re: management capabilites for NFSv4 and requirements draft...

New Message Reply About this list Date view Thread view Subject view Author view Attachment view

From: Robert Thurlow (Robert.Thurlow@eng.sun.com)
Date: 09/30/98-11:52:06 AM Z


Date: Wed, 30 Sep 1998 09:52:06 -0700 (PDT)
From: Robert Thurlow <Robert.Thurlow@eng.sun.com>
Subject: Re: management capabilites for NFSv4 and requirements draft...
Message-ID: <Roam.SIMC.2.0.6.907174326.24615.thurlow@jurassic.eng>

> Sounds like you have never worked with PC-NFS 5.1, 2000 nodes
> across a large state.  I don't thing you realize just how hard it is
> to do this.

My sympathies, but this is a PRODUCT, and not reflective of the
capabilities of the PROTOCOL.  If you'd had a more modern client,
I think you might have been happier, despite the fact that you
might have been using the identical PROTOCOL.

> You are saying that vendors should write a work around for the client.

I'm saying each client implementation should meet the needs of the
user community.  To do this, implementors need a good NFS protocol,
access to name services and a lot of good code.  You seem to think
that NFS must adopt functionality from all other protocols so it
can be the only thing on the client, and I believe this is seriously
misguided.  NFS V4 should subsume its "crutch" protocols (mounting
and locking), but should not go shopping for more functionality to
duplicate gratuitously.

> Upset ON
> 
> I assume by "wire protocol "  you mean that once the file is sent or
> recieved we do not care any more.

No, I mean the RPCs exchanged between client and server as specified
by RFC 1813, the NFS V3 specification.  Our job in this working group
is updating and adding to that spec.

> We are talking about a single function which passes the Platform and
> the User ID, and returns a filename. The server would just lookup
> a table and run a programme which would give the anwser.

I assume this data is available in a name service, and not in files
on the server.  With that true, funneling all requests to the server
just introduces a bottleneck, so it is better that the client refer
to the name services directly.  The name services can be several things:
NIS, NIS+, DHCP, LDAP, etc.  Service Location Protocol and DNS cover
some of this territory, as well.  The LAST thing we need is to try to
add name service calls to NFS to give administrators yet another thing
to configure and manage.

> This protocol must have the functions to support the administrator.
> Such as finding a home directory, redirects,  setting up the NFS
> enviroment, being able to remove locks.

I agree with you only on redirection.  The client should certainly
be told clearly that it should look up the resource again in the
name services, and it might be worth an RPC to find the new location,
as Mike Eisler suggests.

Rob T


New Message Reply About this list Date view Thread view Subject view Author view Attachment view

This archive was generated by hypermail 2.1.2 : 03/04/05-01:46:28 AM Z CST