There are two broad approaches.  Encapsulating PPP in Hyphos (aka
HyPPPhos), or having each cradle be its own PPP server.  Each has
advantages and disadvantages, though I am leaning toward the latter
solution.

HARP = Hyphos ARP

If we encapsulate PPP in Hyphos, we can do the HARP completely on the
gateway machine.  That is, a node can always talk to the base station
without HARP by hardcoding a base station address, and the base
station can keep its own HARP cache based on incoming traffic.  It
still has to nose around inside PPP packets, but not too bad, assuming
we can prevent the Pilot from doing VJ compression.  Rather inelegant
overall, and the basestation becomes a bottleneck.

The nodes (aka cradles) speak PPP directly to the Pilots and IP is
encapsulated in Hyphos, raw, say "HyIPhos".  The issue here is HARP,
but if we make the Hyphos addresses and the IP addresses identical, we
should be in ok shape.