CATNIP

                               Robert Ullmann
                       Lotus Development Corporation
                               29 March 1994


CATNIP Agenda

  o  Agenda bashing
  o  Soapbox
  o  Technical issues
  o  Tuba differences
  o  Questions


Technical Issues

  o  Translating fragments
  o  TTL on ICMP cache setup
  o  TTL conversion to/from IPX "transport control"
  o  Format of IPX addresses
  o  Local IPX numbers


TUBA and CATNIP

  o  Source NSEL in CLNP conversion
  o  TCP and UDP checksums
  o  Addressing plans
  o  Format of PTR RRs in the DNS


Information

  o  draft-ietf-catnip-base-03.txt
  o  ftp.std.com:/pub/catnip
  o  catnip-request@world.std.com
  o  +1 617 693 1315


Soapbox

  o  CATNIP IPng is a new network layer

  o  APIs used by IPv4, CLNP and IPX applications do not change

  o  Network service access point interface used by transports 
     does not change

  o  Subnetwork access (e.g., ES-IS) does not change

  o  New protocols, implementations, and instances of implementations
     can then take advantage of addressing incrementally and 
     independently


Translating fragments

  o  Data Unit ID in IPv4 and CLNP is 16 bits

  o  ID in CATNIP fragment option is 64 + address

  o  If fragmented by IPv4 or CLNP first, use 48 zeros, 16 bit UID,
     source address

  o  If fragmented by CATNIP first, use low 16 bits

  o  If translated multiple times, it works okay most of the time


TTL in ICMP cache setup

  o  Set to 1 to ensure adjacency

  o  Is there a reason why some forwarding configurations might want
     to intentionally use more than 1?

  o  Message sent back to adjacent station on subnetwork


TTL conversion to IPX

  o  IPX counts up from zero

  o  Existing IPX considers 16 as expired

  o  IPX 0 must map to catnip with a reasonably large TTL (64 in 
     current draft)

  o  CATNIP 0 must map to IPX expired

  o  Can we avoid locking IPX into 16?  Can IPX (re-)configure the 
     limit now?


Format of IPX addresses

  o  AFI 47, Novell ICD (tba), network, node address
  o  Followed by 2 zero octets in aligned form
  o  Should a 2 octet field be added somewhere in the middle?
  o  If so, what is it there for?


Local IPX numbers

  o  Local (unregistered) numbers apparently cannot be
     distinguished by format

  o  Systems doing translation are going to work locally
     with unregistered numbers even if we'd like to prohibit it

  o  Encourage Novell to allocate a local-use block?

  o  Does any of this make local use of Appletalk mappings
     workable?


Source NSEL in CLNP

  o  Should the source NSAP selector be zero?
  o  Or should it be a copy of the destination NSEL?
  o  What are the arguments?
  o  No reason why CATNIP and TUBA should be different.


TCP and UDP checksums

  o  CATNIP specifies last 4 octets of NSAP (-SEL)

  o  SIPP does something similar to avoid having to patch up
     transport checksums

  o  TUBA ought to use last 4

  o  When they are some MAC serial number, it is still a very good
     end-to-end check

  o  Future transport protocols (TP4) can do better


Addressing plans

  o  TUBA plan for allocation of NSAPs in the Internet
  o  Other OSI plans
  o  CATNIP mapping of IPv4 and IPX plans
  o  No conflict between these, provided that systems don't try
     to divine remote system capabilities by inspecting the
     address(es) offered


Format of PTR RRs

  o  Forward zone format is agreed (RR is "NSAP")

  o  Inverse zone is rooted at .NSAP

  o  CATNIP specifies octet breaks

  o  Others suggest nibbles, which looks like it will be very
     painful for a very long time

  o  Really a DNS design problem, but unlikely to be fixed by the
     DNS working group


Questions

  o  According to Robert's Rules, a motion to adjourn is in order
     at any time

  o  I understand Seattle has some excellent microbreweries, I
     should have located a few by the time you read this

  o  I am buying (at first)

  o  TUBA is the next session, at 16:00