On 12/24/2021 1:04 PM, Michael StJohns
wrote:
On 12/24/2021 3:02 PM, Christian Huitema wrote:
The collision risk itself depends on the population size. If call signs are used as "pet names", they are only expected to be unique in the small scope of "pets". But if thy are expected to be unique worldwide and forever, then they have pretty much the same design constraints as IPv6 addresses. They can probably be a little shorter because there is no expectation of routability, but I would be surprised if this required fewer than 80 bits -- I would probably go for 96 bits.
If you're looking for statistical uniqueness, this is the birthday collision problem.
If you set the desired chance of collision in the random space at less than 1 in 100,000,000 , 8 octets covers about 600K identities, 10 octets about 155 million and 11 octets about 2.5 billion. Varying the acceptable collision risk up or down assuming everyone is using good quality RNGs with negligible bias changes the number of bits you need to generate to cover an expected space.
https://en.wikipedia.org/wiki/Birthday_problem
n = sqrt (2*d * ln(1/1-p)) where n is the number of expected picks, d is the size the space (e.g. unique identities), and p is the probability of a collision. E.g. p = 10-8, d = 264 gives ~600K for the 8 octets.
(I calculated the above numbers a while back for certificate issuance given random serial numbers).
That's a long winded way to say that 10 octets (96 bits) is probably not enough for a space of trillions as PHB proposes.
PHB is also proposing to record assignments in a ledger, so it is possible to use the ledger to help mitigate collisions. Pick a name, check against the ledger, pick a different one if the first name is already taken. In that case, the desired scope of uniqueness is "all users trying to get a new identity at the same time", which ought to be a much smaller number than trillions.
-- Christian Huitema
