CURRENT_MEETING_REPORT_


Reported by Steve Deering/Xerox Parc, from notes by Jeff Mogul and James
VanBokkelen

MINUTES

This was the second meeting of the Router Discovery (nee Gateway
Discovery) working group.  Jeff Mogul served as acting chair, in
Deering's absence.

The proposed protocol from the December meeting was reviewed.  The
significant features are:

  o It is an ICMP extension.
  o Routers periodically multicast router reports; hosts multicast
    router queries at boot time only.
  o Use of broadcast instead of multicast is permitted but discouraged.
  o A router report does not include a subnet field.
  o A router report includes a holding-time field and a
    preference-level field.
  o In cases where more than one subnet is assigned to the same
    physical network, a router may include multiple addresses (i.e.,
    one for each subnet) in a single router report.

Jeff identified the following open issues:

 1. Meaning of preference levels:
    Nobody seemed to know what to do with more than three levels
    (primary, backup and last chance?).
    Decision:  use 8 or 16 bits, whatever fits in the packet format.
 2. Choice of multicast period:
    Noted that ES-IS uses 10 seconds; we may want slower?
 3. How should router respond to query, unicast or multicast?
    Mike Karels proposed that routers be allowed to attempt to avoid
    unnecessary replies, substituting a single broadcast or multicast
    for several unicast replies.
    Decision:  "keep it simple", i.e., always send unicast replies.
 4. Who writes the RFC?
    No volunteers, so it's still Deering's responsibility.
 5. Do clients dally before sending queries?
    Yes, so that if a LANful of X terminals reboot from ROM at the same
    instant, they don't all hit the broadcast at the same instant.

Other issues raised:

  o Use on non-broadcast media.
    Dismissed as too complicated.
  o Do routers dally before replying?
    Someone suggested that the router dally randomly (mean time based
    on pref level) to avoid overwhelming client.  We argued over

                                  1






    whether the clients needed all the possible router responses right
    off.  However, since we don't want to invent a protocol to stop the
    other N routers from responding, we realized that if we were
    already going to expend the resources for N+1 packets on the wire,
    we might as well try to make use of them at the client.  So,
    dallying seems to be wanted.
  o When to query?
    Drew Perkins proposed a minor shift of definitions to allow initial
    query to be sent when a gateway is first needed (i.e., when first
    sending to an off-subnet destination), rather than at boot time.


ATTENDEES

    Ballard Bare             bare%[email protected]
    Art Berggreen            [email protected]
    Richard Bosch            [email protected]
    Ron Broersma             [email protected]
    John Cavanaugh           [email protected]
    James R. Davin           [email protected]
    Farokh Deboo             [email protected]
    Rich Fox                 [email protected]
    Mike Karels              [email protected]
    Tony Mason               [email protected]
    Keith McCloghrie         [email protected]
    Bill Melohn              [email protected]
    Jeff Mogul               [email protected]
    John Moy                 [email protected]
    Drew Perkins             [email protected]
    Michael Petry            [email protected]
    Mark Rosenstein          [email protected]
    Tim Seaver               [email protected]
    Tony Staw                [email protected]
    James VanBokkelen        [email protected]
    John Veizades            [email protected]
    Steve Willis             [email protected]
    John M. Wobus            [email protected]
    David Paul Zimmerman     [email protected]



                                  2