InfoMagic Standards 1994 January/InfoMagic Standards - January 1994.iso/inet/ietf/fddi/90jul.min

Text File | 1993-02-17 | 4KB | 113 lines

CURRENT_MEETING_REPORT_



Reported by Richard Fox/Synoptics

FDDI Minutes

The meeting was solely comprised of a presentation by Caralyn Brown and
Doug Bagnall called, ``ARP extensions for Dual Mac Stations''.

Currently ARP supports a 1-1 mapping of IP addresses to MAC addresses.

FDDI supports the notion of 1-2 mapping of IP addresses to MAC
addresses.

Our goal is not to have a TCP connection break when a wrap happens.  To
meet this objective it was suggested that an extension to the current
ARP protocol is needed, where the new ARP protocol supplies more than a
1-1 mapping but a 1-many mapping.  An example of this is:

      ARP response= <ip><mac1,ring1><mac2,ring2>

One step identified in achieving this is to add a new SNAP value.

At this point 2 approaches were presented and compared.

Solution 1:  Hybrid approach

Have a parameter that says that no backward compatibility is to be
maintained.  Thus, send old style ARP but encode stuff in target fields.

Advantages:  only need to send 1 ARP for all cases.  Disadvantages:
encoding may break some implementations and this solution doesn't scale
very well.

Some people said that this method is better solved at layer 3; reply to
this was to rewrite layer 3; thus this solution is less radical than
rewriting layer 3.

Solution 2:  Extended ARP

This solution requires that a new ARP packet be sent out each interface
(this packet is called an EARP and is slightly different than the normal
ARP packet).  After an EARP is sent the station must set a timer and
wait for a response.  If no response is received then the station must
assume that the receiver of the ARP doesn't understand EARPs and so it
must send out a normal ARP.

Advantages:  backwards compatibility.  Disadvantages:  may need to send
out 2 ARP requests before an answer is received.

                                   1






Other issues that came up with this solution are:

   o When ring wraps/unwraps stations should send ARP to itself to
     update everybody's ARP table -- do this only after a settling
     period.  Some people felt that the SRF frame takes care of this,
     others not convinced, no resolution.
     At this time we listed advantages of allowing stations to have 2
     macs.  The 3 identified reasons are:

      -  Load balancing (transparent).
      -  Transparent error recovery.
      -  Dual mac in wrap:  you don't know where response came from.

   o Need EARP since non-wrapped stations can use wrong ring when a
     station is wrapped.  EARPs keeps effect to wrapped stations
     only.(??)  At this point we got into varied discussions on how
     wrapped rings and IP do not get along.  Some people want to force
     all single MAC stations to be connected to the primary ring only
     (or at least on the same ring), others feels that this rule breaks
     the concept of FDDI.
   o It was suggested that we continue to use RFC 1122 for ARP cache
     handling.

Attendees


Douglas Bagnall          [e-mail]
Alison Brown             alison@[e-mail]
Caralyn Brown            [e-mail]
Cho Chang                [e-mail]
Andrew Cherenson         [e-mail]
Cyrus Chow               [e-mail]
Paul Ciarfella           [e-mail]
Nadya El-Afandi          [e-mail]
Richard Fox              sytek![e-mail]
Michael Grobe            [e-mail]
Susan Hares              [e-mail]
Peter Hayden             [e-mail]
Ajay Kachrani            [e-mail]
Jay Kadambi              [e-mail]
John LoVerso             [e-mail]
Rebecca Nitzan           [e-mail]
James Reeves             [e-mail]
Bill Townsend            [e-mail]
Bert Williams            [e-mail]
Linda Winkler            [e-mail]
Sijiam Zhang             [e-mail]



                                   2