CURRENT_MEETING_REPORT_



Reported by Peter Honeyman/University of Michigan



SYNOPSIS

Trans- and inter-continental distributed file systems are upon us.  The
consequences to the Internet of distributed file system protocol design
and implementation decisions are sufficiently dire that we need to
investigate whether the protocols being deployed are really suitable for
use on the Internet.  There's some evidence that the opposite is true,
e.g., some DFS protocols don't checksum their data, don't use reasonable
MTUs, dont offer credible authentication or authorization services, dont
attempt to avoid congestion, etc.

Accordingly, a working group on DFS has been formed by the IETF. The WG
will attempt to define guidelines for ways that distributed file systems
should make use of the network, and to consider whether any existing
distributed file systems are appropriate candidates for Internet
standardization.

The WG will also take a look at the various file system protocols to see
whether they make data more vulnerable.  This is a problem that is
especially severe for Internet users, and a place where the IETF may
wish to exert some influence, both on vendor offerings and user
expectations.

dfs-wg@citi.umich.edu is a mailing list for ongoing discussions of the
WG; administrative matters, such as requests to be added or dropped from
the list, should be addressed to dfs-wg-request@citi.umich.edu, not to
the list as a whole.

MINUTES

The meeting was chaired by Peter Honeyman.  At the meeting, plans were
made to meet the following objectives.

OBJECTIVE: Produce a document for implementors and administrators, in
the style of the Hosts Requirements RFCs.

Issues to be addressed include recommendations to be followed when UDP
is used as the transport layer.  Most of these recommendations come from
experiences with TCP. The recommendations include:

   o the use of the transport-layer checksum;
   o techniques for congestion avoidance;
   o techniques for fragmentation avoidance;
   o retransmission strategy based on measured round-trip times.

                                   1






The group intends to identify other recommendations and to flesh out the
details in time for a review at the next IETF meeting.

OBJECTIVE: Standard for Kerberos authentication for NFS.

Several groups have deployed or are preparing to deploy NFS integrated
with Kerberos.  Among these are MIT, U Michigan, and Transarc, Inc..
These groups will describe the protocols they now use for establishing
and maintaining Kerberos credentials in an NFS session.  The intent is
to agree on a common protocol, which will be described in an RFC.

Representatives from MIT, Michigan, and Transarc agreed to describe
their protocols in the dfs-wg mailing list.  At the next meeting of the
IETF, substantive differences between the protocols will be discussed.

OBJECTIVE: Establish the requirements for Internet-friendly DFS
protocols.

DFS protocols that were developed for a LAN environment can behave
abysmally on a WAN. A well designed DFS will balance its performance
needs with those of other users and uses of the network.

Many of the issues concerning the design of DFS protocols depend on one
another, or on advances in other areas under study by the IETF. A
partial list of the areas in which recommendations can be made includes:


   o Naming
   o Data representation
   o Type management
   o Locking
   o Impact of design choices:
      -  Statelessness
      -  Cache management
      -  Choice of transport
   o Use of MTU discovery
   o Authentication and authorization
   o Trusted vs.  untrusted client
   o Time protocol
   o User expectations


The first task is to establish concrete goals to guide the WG in this
area.

GOALS FOR NEXT IETF MEETING

``Guidelines for DFS Administrators and Implementors'' in draft form.
Current status of Kerberized NFS implementations on paper.  Further
discussion on "Guidelines for DFS Designers."



                                   2






ATTENDEES

Richard Basch                         probe@mit.edu
Dave Borman                           dab@cray.com
Peter Honeyman                        honey@citi.umich.edu
Mike Karels                           karels@berkeley.edu
Ole Jacobsen                          ole@csli.stanford.edu
Dan Jordt                             danj@washington.edu
Loius A. Mamakos                      louie@trantor.umd.edu
Tony Mason                            mason@transarc.com
Matt Mathis                           mathis@pele.psc.edu
Leo J. McLaughlin                     ljm@twg.com
Greg Minshall                         minshall@kinetics.com
Don Morris                            morris@ucar.edu
Drew Perkins                          ddp@andrew.cmu.edu
Joel Replogle                         replogle@ncsa.uiuc.edu
Dean Throop                           throop@dg-rtp.dg.com
A. Lee Wade                           wade@orion.arc.nasa.gov
Dan Wintringham                       ydanw@osc.edu



                                   3