RESEARCH: Slashing incentives for validator decentralization

Conversation Validation
1

Decentralization Incentivized Slashing models

This is active area of research for @sunnya97 and myself.

Goals:

  • Validators that are a large fraction of the total voting power need be slashed at high levels. For instance, a validator that was 1% of the network who equivocates could be slashed 5% but 10% validator might get slashed at 10%
  • To discourage large validators from creating sybils, we need to introduce slashing on correlated behavior. A similar fault that occurs across multiple keys in the same slashing period should result in larger amplification factor than
  • Introduce a reward model that demonstrates adversarial control of a consensus key or keys and works within the above.

These effects should encourage delegators to pursue a portfolio strategy with validators/staking pools and incentivize decentralizing the consensus while still encouraging delegators to strongly consider the security of the their validators.

Details:

  1. Slashing formula for a single validator relative to total stake
  2. We need to figure out how correlations work. The validator will be tombstoned on the first fault reported but additional slashing will need to occur if new correlated faults are found. If 10 1% validators fault together the total slashed much exceed 1 10% validator according to the fault in 1.
1 Like
2

I think it’s effectively impossible to create incentives against large self delegations.

This is mostly about holders that don’t run their own or closely associated validators and encouraging a portfolio strategy.

We also need to figure out a way to conduct experiments with this system.

3

This will also require a validator control over delegation mechanism built.

Let’s say, a validator has a decent amount of atoms delegated to him/her and don’t want to accept any new delegations in order to be below increased slashing % risk. The only control present now is commission fee that could be increased but it wouldn’t be sufficient.

4

I know you’re talking about slashing incentives here for decentralization but what if you incentivized Delegators to diversify for decentralization?

Right now there are theories of there being benefits to diversifying, but little to no actual evidence of the benefit in diversifying.
This is because there’s no difference between systematic versus unsystematic risk, since most Validators seem to be doing fine and dandy with regards to uptime.
I have had a theory since mainnet launch that all Validators are doing fine now because there’s no attacks because there’s no actual market price or market value attached to attacking. So this might be a short-term problem once transfers are enabled and we see real exchanges trading Atoms.

But if that theory is wrong, and centralization of Atoms among a few Validators is still present well after transfers are enabled, I’d suggest you create an additional staking rewards formula so that diversified Delegators receive “extra rewards” on top of the normal staking rewards for Delegating Atoms to a diversified group of Validators. Example: if the Atoms in your wallet are diversified more or less equally to at least 50 active Validators then you get 50% extra staking rewards.

Initial questions:
-What’s a “diversified group of Validators” look like?
Hard to say, but a perfectly distributed Delegator wallet with 100 Atoms in the wallet would have 1 Atom to all 100 active Validators. A sliding scale for these extra rewards would also make sense; ie- an ideal diversified wallet would have to see the highest “extra rewards” possible but a somewhat diversified wallet would have to see only somewhat “extra rewards”.

-How does someone “game” this system?

Validators with large self-stakes might charge higher than normal commissions, knowing that Delegators trying to get extra rewards would be essentially forced to delegate to them and pay that higher commission.

Could cause Validators to start to spread out their staked Atoms and take up multiple active Validator spots in order to artificially allow Delegators to falsely “diversify” across multiple active Validators. This could cause a good chunk of smaller Validators to fall out of the active Validator spot they’re currently in.

I’m sure there are more ways to game it, but those are just a couple that come to mind.

-How high would the “extra rewards” have to be in order for Delegators to want to do it?

In my mind, as a Delegator, it would have to obviously compensate me for staking to a lot of Validators who I otherwise wouldn’t normally stake to. And the various reasons for not staking to certain Validators are if:
-I don’t know a thing about the Validator;
-the Validator charges what I think is an irrationally high commission;
-the Validator has had poor performance and possibly might be a slashing risk;
-the Validator is a company or individual I just don’t want to support for personal reasons.
So to quell these factors it would truly have to be a large chunk. Off the top of my head I’m thinking like an additional +50% of whatever the going staking reward is (so if staking rewards are at 10%, this would have to be 15%).

Thoughts?

2 Likes
5

Higher slashing levels for a validator with higher stake impose negative externalities. This can lead to unintentional dynamics and attack vectors :

  • An additional delegation would increase the slashing level for every other delegation and the self-bond stake.

  • It would discourage validators to have a high self-bond stake as the slashing levels increase with every additional delegation.

  • There is a conflict of interest between validators and their delegators as a validator may profit from additional delegations while delegators are harmed by additional delegations.

  • A new delegator with a significant amount would increase the slashing level and and old delegator with a possibly higher amount could decide to leave. The total delegation amount could be lower afterwards. This scenario could also be constructed as an attack on a certain validator with the addition of redelegating afterwards.

  • Every delegator spend time and effort evaluating a validator and optimally trust and partnership emerges. Changing slashing levels could force a delegator to redelgate from a trusted validator and therefore create additional costs.

To avoid these negative externalities i think of a slashsing system that simply applys different slashing conditions on different delegators. For example we could order delegations to a specific validator by a timeline and compute indiviudal slashing conditions on basis of the cumulative total delegation amount. The higher the cumulative total delegation amount the higher your indiviudal slashing risk. Some Advantages:

  • The slashing conditions do not change for existing delegators but still discourage every new delegation. There are no negative externalities

  • New validators without reputation have a chance of getting delegations because the slashing risk is relatively low for new delegators.

  • It encourages partnerships and trust between validators and delegators. Note that you position in the timeline could advance as people that are before you on the line leave and therefore you have lower slashing risk.

Note that this does not adress the problem of creating sybils.

6

I really like the idea. However there are some problems when it comes to implementing i think :

  • How do you set up such a system with variable rewards while maintaining a fixed inflation rate?
  • You claim your rewards form multiple single validators. How should the system take into account that your funds are well distributed and pay you accordingly ?