Chris Belcher [ARCHIVE] on Nostr: 📅 Original date posted:2019-08-06 📝 Original message:On 06/08/2019 02:51, Leo ...
📅 Original date posted:2019-08-06
📝 Original message:On 06/08/2019 02:51, Leo Wandersleb via bitcoin-dev wrote:
> On 8/6/19 7:04 AM, Chris Belcher via bitcoin-dev wrote:
>> However, there _is_ a cost to being a sybil attacker. If we define
>> honest makers as entities who run just one maker bot, and dishonest
>> makers as entities who run multiple maker bots, then we can say that
>> running a dishonest maker operation requires a sacrifice of fee income,
>> because someone doing that would earn more money if they ran an honest
>> maker instead. This happens because of the quadratic V^2 term in the
>> formula calculating the fidelity bond value, which provides this
>> incentive for lumping together fidelity bonds. This V^2 is probably the
>> most important part for privacy.
>
> As established above, there will emerge a market to lock coins, so these locks
> will be readily available without having to buy them. Even with V^2 there is no
> reason to amass more coins beyond a certain point. Running the biggest 5 V^2
> scores should be pretty solid to get in on many coin joins.
We can be much more exact than saying makers get in on "many" coins. The
supporting document "Financial mathematics of JoinMarket fidelity bonds"
contains calculations for exactly this:
https://gist.github.com/chris-belcher/87ebbcbb639686057a389acb9ab3e25b#sybil-attacks-from-enemies-within
The document finds that with realistic real-world data, the makers with
the top 5 most valuable bonds will be chosen 48% of the time. So
approximately half:half success for one coinjoin. This isn't enough to
deanonymize every single coinjoin. For example, the tumbler script by
default makes around 16 transactions so the odds of a successful sybil
attack is (0.48)^16 = 8 parts per million, with the success probability
reducing exponentially after each additional coinjoin.
>> Another way is to require the bond signature proofs to involve the
>> one-time taker identifier, and so be different every time. This
>> basically requires fidelity bond privkeys to be online in hot wallets,
>> and so should massively increase the difficulty of renting TXOs because
>> the maker and the TXO owner need to be in constant real-time communication.
>
> Requiring the bond to reside on a hot wallet would be a massive disadvantage.
Hopefully it won't come to that and we can invent some other way to stop
renting TXOs. But if that's the only way then we'd have to code it in
order to protect the interests of takers.
The most dangerous source of rented TXOs seems to be the coin age form
of fidelity bond. Hodlers could have coins already in a hardware wallet
or cold storage and just sign proofs renting their UTXOs to earn an
extra income without changing their setup at all. Bonds from OP_CLTV and
OP_RETURN burned coins seems to me a much less likely source of rented TXOs.
Because of that, it seems to me only coin age fidelity bonds would be
required to be on hot wallets.
Another option worth considering is the have a separate lower interest
rate for coin age bonds compared to OP_CLTV bonds, this would reflect
the lower sacrifice for coin age (past sacrifices must be worth less
than future sacrifices, because of risk and uncertainty of the unknown
future, as well as the risk of rented UTXOs)
> No matter how you look at the whole problem of sibyl attacks, the honest maker
> will have operational costs and gain fees and the sibyl attacker will have the
> same plus profit from the deanonymization. As long as makers hunt marginal
> profits, the sibyl attacker having the higher margin from deanonymization will
> always win. The fidelity bonds would make this even worse, as increased
> complexity and entry cost would not favor more makers but less even before the
> centralization incentive mentioned above (V^2). To say that old holders have
> bitcoins laying around that they can use for such bonds is a fallacy as they
> could just as well rent them out on a bonds market.
I think this is absolutely wrong, because sybil attackers give up some
fee income. Here is a worked example:
Let's say the sybil attacker is operating the top 5 most valuable maker
bots. If this attacker has X coins they would split them equally into 5,
so each maker has X/5 coins and their bond is worth (X^5)^2 = X^2/25,
with a total of 5 bots the fee income would be proportional to 5*X^2/25
= X^2/5. However if an honest maker had X coins they could create a
single bond which would be worth simply X^2 with a fee income
proportional to X^2. So the honest maker has a fee income higher by a
factor of 5 than the sybil attacker. The sybil attacker must take a 5x
hit to their fee income in order to sybil attack. This is the crucial
effect of the V^2 term.
The V^2 term is important, it just has the downside of incentivizing
renting of coins. If we can make that impossible then the problem would
go away.
> How about turning this upside down and shift the incentives from being taker to
> being maker by introducing a mandatory fee? If each join costs 1% per maker,
> people would initially gasp and reject to update to that version but those who
> do, will do to become makers, increasing the maker count massively and
> eventually most people in frequent need of joining will also become makers to
> offset the costs of being takers.
>
> With these changed rules again the sibyl attackers would still have their
> competitive edge and would flood the market with even more cheap offers but now
> everybody would have an incentive to do the same and as makers have to have the
> UTXOs, it's not free to sibyl attack already.
>
Apart from the inability of developers to enforce any kind of price, I
don't think this scheme would fix the sybil attack problem, because a
sybil attacker still gets a higher gain (deanonymization + fees)
compared to honest makers (who earn just fees)
📝 Original message:On 06/08/2019 02:51, Leo Wandersleb via bitcoin-dev wrote:
> On 8/6/19 7:04 AM, Chris Belcher via bitcoin-dev wrote:
>> However, there _is_ a cost to being a sybil attacker. If we define
>> honest makers as entities who run just one maker bot, and dishonest
>> makers as entities who run multiple maker bots, then we can say that
>> running a dishonest maker operation requires a sacrifice of fee income,
>> because someone doing that would earn more money if they ran an honest
>> maker instead. This happens because of the quadratic V^2 term in the
>> formula calculating the fidelity bond value, which provides this
>> incentive for lumping together fidelity bonds. This V^2 is probably the
>> most important part for privacy.
>
> As established above, there will emerge a market to lock coins, so these locks
> will be readily available without having to buy them. Even with V^2 there is no
> reason to amass more coins beyond a certain point. Running the biggest 5 V^2
> scores should be pretty solid to get in on many coin joins.
We can be much more exact than saying makers get in on "many" coins. The
supporting document "Financial mathematics of JoinMarket fidelity bonds"
contains calculations for exactly this:
https://gist.github.com/chris-belcher/87ebbcbb639686057a389acb9ab3e25b#sybil-attacks-from-enemies-within
The document finds that with realistic real-world data, the makers with
the top 5 most valuable bonds will be chosen 48% of the time. So
approximately half:half success for one coinjoin. This isn't enough to
deanonymize every single coinjoin. For example, the tumbler script by
default makes around 16 transactions so the odds of a successful sybil
attack is (0.48)^16 = 8 parts per million, with the success probability
reducing exponentially after each additional coinjoin.
>> Another way is to require the bond signature proofs to involve the
>> one-time taker identifier, and so be different every time. This
>> basically requires fidelity bond privkeys to be online in hot wallets,
>> and so should massively increase the difficulty of renting TXOs because
>> the maker and the TXO owner need to be in constant real-time communication.
>
> Requiring the bond to reside on a hot wallet would be a massive disadvantage.
Hopefully it won't come to that and we can invent some other way to stop
renting TXOs. But if that's the only way then we'd have to code it in
order to protect the interests of takers.
The most dangerous source of rented TXOs seems to be the coin age form
of fidelity bond. Hodlers could have coins already in a hardware wallet
or cold storage and just sign proofs renting their UTXOs to earn an
extra income without changing their setup at all. Bonds from OP_CLTV and
OP_RETURN burned coins seems to me a much less likely source of rented TXOs.
Because of that, it seems to me only coin age fidelity bonds would be
required to be on hot wallets.
Another option worth considering is the have a separate lower interest
rate for coin age bonds compared to OP_CLTV bonds, this would reflect
the lower sacrifice for coin age (past sacrifices must be worth less
than future sacrifices, because of risk and uncertainty of the unknown
future, as well as the risk of rented UTXOs)
> No matter how you look at the whole problem of sibyl attacks, the honest maker
> will have operational costs and gain fees and the sibyl attacker will have the
> same plus profit from the deanonymization. As long as makers hunt marginal
> profits, the sibyl attacker having the higher margin from deanonymization will
> always win. The fidelity bonds would make this even worse, as increased
> complexity and entry cost would not favor more makers but less even before the
> centralization incentive mentioned above (V^2). To say that old holders have
> bitcoins laying around that they can use for such bonds is a fallacy as they
> could just as well rent them out on a bonds market.
I think this is absolutely wrong, because sybil attackers give up some
fee income. Here is a worked example:
Let's say the sybil attacker is operating the top 5 most valuable maker
bots. If this attacker has X coins they would split them equally into 5,
so each maker has X/5 coins and their bond is worth (X^5)^2 = X^2/25,
with a total of 5 bots the fee income would be proportional to 5*X^2/25
= X^2/5. However if an honest maker had X coins they could create a
single bond which would be worth simply X^2 with a fee income
proportional to X^2. So the honest maker has a fee income higher by a
factor of 5 than the sybil attacker. The sybil attacker must take a 5x
hit to their fee income in order to sybil attack. This is the crucial
effect of the V^2 term.
The V^2 term is important, it just has the downside of incentivizing
renting of coins. If we can make that impossible then the problem would
go away.
> How about turning this upside down and shift the incentives from being taker to
> being maker by introducing a mandatory fee? If each join costs 1% per maker,
> people would initially gasp and reject to update to that version but those who
> do, will do to become makers, increasing the maker count massively and
> eventually most people in frequent need of joining will also become makers to
> offset the costs of being takers.
>
> With these changed rules again the sibyl attackers would still have their
> competitive edge and would flood the market with even more cheap offers but now
> everybody would have an incentive to do the same and as makers have to have the
> UTXOs, it's not free to sibyl attack already.
>
Apart from the inability of developers to enforce any kind of price, I
don't think this scheme would fix the sybil attack problem, because a
sybil attacker still gets a higher gain (deanonymization + fees)
compared to honest makers (who earn just fees)