Towards even delegation distribution

[robert-zaremba]

Blockchain is all about decentralization.
In Tendermint, users have a full power to decide if and where to delegate their stake. Going to the first "cheapest" validator in the list will lead to a network centralization. In theory validators should "fight" for the delegates, offering some incentives beyond low commission. However we don't have a good balancing force to "drive" for a good validator decentralization.

Let's find potential solutions.

@anilcse has a data outlining the validators centralization.

cc: @sunnya97 @glandua

[robert-zaremba]

Tax

I was sharing this idea in few places. Originally, I believe, it was researched by Cardano.
Whenever a stake goes x% above average (let's say 150% of average), the validator is taxed lineary. The tokens goes either to validators with lower stake (eg 50% of average) or a special pool governed by the gov /DAO or a combination of the latter two.

[robert-zaremba]

Hard limit

@anilcse was proposing an implementation of MaxValidatorRewardShare gov param. It would dependent on a number of active validators and total staked amount. This would force delegators to chose lower end validators over top (highest staked) validators for thier delegations. Otherwise they would miss potential rewards (low shares vs amount staked).

[alexanderbez]

I'm sure @sunnya97 has some opinions on this. In fact, we had this very discussion years ago before the Hub went live.

[robert-zaremba]

ping @sunnya97

[zmanian]

IMO staking derivatives are the best known mechanism to incentive stake decentralization. Fungible pools on top of staking derivatives will want to limit their exposure to any one validator.

[robert-zaremba]

Staking derivatives is a good approach, but I'm not sure if this will really prohibit centralization. Delegation is already a decentralized process, and if delegators don't put enough effort for a right decentralization, then adding one more governance layer (of choosing delegation shares for staking derivatives) won't help.

[anilcse]

One question I encountered when proposing MaxValidatorRewardShare was about operating multiple validators (applies for Tax suggestion too). Practically anyone can operate any number of validators. We can think of some governance procedure to elect validators or remove in such cases (it has other downsides as well).

[anilcse]

But then you would need to operate as:

• <your-name>-2 -- we should communicate that this is a bad practice for delegators.
• an unknown validator -- which should discourage delegators

But exchanges can operate multiple validators without expecting external delegations. That's a problem

[robert-zaremba]

ah, I see. I thought it "normally" doesn't make sense to use multiple validators by the same entity.
BTW: What's a reason for running multiple validators by the same entity?

[zmanian]

Polkadot limits the amount of stake that any single validator can have

[anilcse]

true, but if we consider binance for example, they can just operate multiple validators. Stake is still centralized if we consider entities. With years of observations, we see no value from exchanges in governance or community activities. One possible solution for this could be, keeping a max limit on stake or rewards + allowing governance to remove a validator from validator set.

[robert-zaremba]

allowing governance to remove a validator from validator set.

I like that!

[hxrts]

Ideas I've seen in this direction:

• Liquid staking
• Massively increasing slashing (see takeaways section here)
• Correlated slashing
• Proportional slashing

[robert-zaremba]

Liquid staking

This requires a well managed delegation process on the liquid staking protocol. Do we have that implemented?

[hxrts]

Lido for Terra https://docs.terra.lido.fi/ in which the Lido DAO manages delegations, so comes with its own issues.

[anilcse]

Since Liquid Staking is also not solving this problem at core, should we atleast create a relationship b/w self-stake -> max delegation requirement?

[anilcse]

One more approach i am thinking about is, we should introduce a base rewards param for validators. Like 20% let's say. 20% rewards are distributed equally among all the validators and only remaining 80% will be distributed according to the delegations.

[alexanderbez]

As opposed to the stake-weighted approach we use today?

[anilcse]

Yes