This seems unstable to me. The IP generator would need its own random
state to avoid being desynchronized by other things that need random
numbers, like acknowledgement numbers. Two runs could get out of sync
after the first scan group, if a different number of retransmits were
Even if the IP generator has its own state, you could get different
results across versions because of different reserved IP ranges.
That's not so bad, because any IP address that became reserved would
be omitted in the newer version and replaced by a different IP at the
end of the list. But it means you couldn't count on the same seed
giving you the exact same IP addresses.