trying to understand SD-NAT

[Francis Dupont]

The mapping function in 4.5 is pretty bad and if one wants to
subscriber port ranges to ranges it is far simpler to allocate
directly ranges to subscriber, i.e., if [1024, 1024+N] is mapped
to [P, P+N] than [P, P+N] could be assigned in place of [1024, 1024+N].

So the mapping function should be more complex. There was some research
about this for stateless A+P, something which seems to work well
is for instance for the range 1024+n n=0..N with N < 1024 to map
for the subscriber S to (n << 9) | (S << 3) | (n & 0b111)
(this is the function I'd like to implement).

Regards

Francis Dupont <fdupont at isc.org>