Validator nodes are online servers running a blockchain's client software. They all keep a copy of the ledger and are constantly talking to each other to make sure the copies are consistent as new data is added.
The fact that the ledger is not under control of a single entity but distributed among a network of independent validators is what makes blockchains:
- Require less trust than traditional options.
- Censorship resistant.
- Byzantine fault tolerant.
Validators agree on the state of the ledger using a consensus algorithm that varies for each blockchain. While the implementation of the groundbreaking FCP protocol is being completed, Flare is currently using a variant of Avalanche's Snowman++ algorithm.
In Avalanche's Snowman++, each round a validator is randomly selected to act as the leader and propose new blocks to be added to the ledger, which are then validated by the rest of nodes. To provide Sybil resistance, the probability of a node's being elected the leader is proportional to the node's stake, so this is effectively a Proof-of-Stake consensus. The shortcomings of Proof-of-stake are well known and include risk of centralization and the rich-get-richer effect, for example.
To compensate for this, Flare's version of Snowman++ reduces the importance of a node's stake and introduces a meritocratic factor: All Flare validators are also FTSO Data Providers, so their performance in this role has an impact on their chance to become round leaders.
More precisely, the probability \(P\) of being the leader depends both on a node's stake (\(VotePower\)) and on its performance as a Data Provider (evaluated through its \(RewardRate\)) like this:
And then normalized so the probabilities for all nodes add up to \(1.0\).
As it can be seen:
- The logarithm (which might be replaced by a square root or similar monotonically decreasing function) reduces the importance of large stakes.
- Multiplying by the \(RewardRate\) benefits the nodes that consistently provide good FTSO data.
Both the \(VotePower\) and the \(RewardRate\) are evaluated once a week based on the previous week's results.
The plots below show the equalizing effect of this formula on actual data taken from flaremetrics.io:
As it is readily apparent, the distribution to the right is far more egalitarian, while still rewarding high stakes and FTSO performance.