Major update in EC and Secret Leader Election Specification #1054
Conversation
|
|
||
| This is modelled via a Binomial distribution with parameter`p=ExpectedLeaderPerRound/TotalPower` and `n= MinerPower`. | ||
|
|
||
| For ease of implementation, we use the the Poisson distribution with parameter `lambda = MinerPower*ExepectedLeader` to approximate this Binomial distribution. |
There was a problem hiding this comment.
I wouldn't say Poisson is an approximation for ease of implementation. I initially thought Binomial(ExpectedLeadersPerRound, MinerPower/TotalPower) was the "correct" or ideal distribution, but after much discussion with @jzimmerman, I came to realize this: since every miner is rolling uncorrelated/independent dice, and we want to be fully invariant to miner pooling (not just in expectation), Poisson is actually the ideal distribution.
The table below might still be worth presenting, but the framing should be something like "you might think binomial is better, but it isn't, but look, it doesn't matter much anyway" rather than "our approximation is good enough".
| - Compute the probability distribution of winning `k` blocks (i.e. Poisson distribution see below for details) | ||
| - Let `h_n` be the normalised VRF, i.e. `h_n = vrfout/2^256` | ||
| - The probability of winning one block is `1-P[X=0]`, thus if `h_n` is less than `1-P[X=0]`, the miner wins at least one block. | ||
| - Similarly if `h_n` is less than `1-P[X=0]-P[X=1]` we have at least two blocks and so on. |
There was a problem hiding this comment.
I might add: "While it is not permitted for a single miner to publish two distinct blocks, in this case, the miner produces a single block which earns two block rewards."
|
|
||
| Filecoin is building on the principle that a miner possessing X% of network power should win as many times as X miners with 1% of network power in the election algorithm. | ||
|
|
||
| This is modelled via a Binomial distribution with parameter`p=ExpectedLeaderPerRound/TotalPower` and `n= MinerPower`. |
There was a problem hiding this comment.
If you want to pick an intuitive Binomial model to compare to, IMO the most intuitive is n=ExpectedLeadersPerRound and p=MinerPower/TotalPower
This PR is applying all the latest changes in the Expected Consensus, Leader Election and the rationale behind Poisson Sortition of
WinCount. It removes all mention of Election PoSt which was very out-of-date.