Re: [bitcoin-dev] MAD-HTLC

2020-06-23 Thread Stanga via bitcoin-dev
Of course the order at the end should have been switched:

Consider first the case where Alice *does not* publish preimage "A": Bob
can safely publish preimage "B" and get both the Deposit and Collateral
tokens after the timeout.
Now, consider the case where Alice *publishes* preimage "A": If Bob
publishes preimage "B" he gets nothing (and so does Alice - this is the
mutual assured destruction), and if he doesn't, he gets the Collateral
tokens.


On Tue, Jun 23, 2020 at 3:47 PM Stanga  wrote:

> Hi ZmnSCPxj,
>
> Thank you for taking the time to respond, these are very good points.
> Responses inline.
>
> On Tue, Jun 23, 2020 at 12:48 PM ZmnSCPxj  wrote:
>
>> Good morning Itay, Ittay, and Matan,
>>
>> I believe an unstated assumption in Bitcoin is that miners are
>> short-sighted.
>>
>> The reasoning for this assumption is:
>>
>> * Deployment of new mining hardware controlled by others may occur at any
>> time you do not control.
>>   * Thus, any transactions you leave on the table are potentially taken
>> by somebody else and not by you.
>>   * Sudden changes in hashpower distribution may reduce your expected
>> future earnings, so any future theoretical earnings should be discounted
>> (*in addition to* expected return-on-investment on getting money you can
>> invest *now*).
>>
>
> Our analysis assumes constant difficulty, i.e., no significant changes of
> the miners set. Indeed, hash-rate changes typically occur at a much larger
> granularity than your average HTLC timeout. For instance, we noticed plenty
> of lightning nodes use timeouts of a day. So, we do not consider
> optimization at infinity, just a day ahead, and within this time frame all
> the factors you mentioned are not expected to dramatically change.
>
> That being said, it would be interesting to analyze the effect of miners
> joining during the HTLC duration. Intuitively, this shouldn’t affect the
> results, as those new miners have the same incentive to wait for the
> higher-paying tx.
>
>
>>
>> It also strikes me that, in a world with RBF and CPFP, the same endpoint
>> (i.e. miners earn the entire fund of the HTLC) is achieved by existing
>> HTLCs, without the additional branch and script opcodes needed by MAD-HTLC.
>> For example, if an HTLC is confirmed but the hashlock-claiming
>> transaction is not being confirmed (because miners are holding it up
>> because Bob is offering a much higher fee in the future for the
>> timelock-claiming transaction), then Alice can, regardless of the reason
>> why it is not being confirmed, bump up the fee with RBF or CPFP.
>>
>> If the fee bump offered by Alice is sufficiently large, then miners will
>> start re-preferring the Alice hashlock transaction.
>> To counter this, Bob has to bid up its version higher.
>>
>> As the timeout approaches, Alice can bump up its fee until it is just 1
>> satoshi short of the total fund.
>> It is rational for Alice to do so since at timeout, it can expect to lose
>> the entire fund.
>> In order for Bob to win, it has to beat that fee, at which point it
>> equals or exceeds the total fund, and miners get the total fund (or more).
>>
>> Knowing this end-point, rational Bob will not even begin this game.
>>
>> I think this research considers these two endpoints to be distinct:
>>
>> * Bob misbehaves and the entire fund is punished by miners, leaving
>> miners with the fund and Alice and Bob without money (MAD-HTLC).
>> * Bob misbehaves, Alice counters, and the ensuing fee war leads to fees
>> approaching the fund value, leaving miners with the fund and Alice and Bob
>> without money (standard HTLC).
>>
>> But in practice I think both endpoints are essentially equivalent.
>>
>
> These are not the same scenario, since in HTLC there is a race between
> Alice and Bob. Alice might not wish to pay the full HTLC amount once she
> sees Bob is trying to cheat. She could wait until close to the timeout so
> as to reduce the time Bob can respond. Of course Bob would do the same. So
> this is an actual race, and Bob takes no risk since his payment is all from
> the HTLC amount. Mutual destruction is only assured under certain
> assumptions in HTLC. MAD-HTLC achieves security without relying on such
> assumptions.
>
>
>>
>> --
>>
>> What MAD-HTLC can do would be to make different claims:
>>
>> * Inputs:
>>   * Bob 1 BTC - HTLC amount
>>   * Bob 1 BTC - Bob fidelity bond
>>
>> * Cases:
>>   * Alice reveals hashlock at any time:
>> * 1 BTC goes to Alice
>> * 1 BTC goes to Bob (fidelity bond refund)
>>   * Bob reveals bob-hashlock after time L:
>> * 2 BTC goes to Bob (HTLC refund + fidelity bond refund)
>>   * Bob cheated, anybody reveals both hashlock and bob-hashlock:
>> * 2 BTC goes to miner
>>
>> This is an actual improvement over HTLC: Bob misbehavior leads to loss of
>> the fidelity bond.
>> The above cases can be assured by requiring both Alice and Bob to sign in
>> the alice-hashlock branch, so that the splitting of the fund is enforced,
>> and SegWit signing 

Re: [bitcoin-dev] MAD-HTLC

2020-06-23 Thread Stanga via bitcoin-dev
Hi ZmnSCPxj,

Thank you for taking the time to respond, these are very good points.
Responses inline.

On Tue, Jun 23, 2020 at 12:48 PM ZmnSCPxj  wrote:

> Good morning Itay, Ittay, and Matan,
>
> I believe an unstated assumption in Bitcoin is that miners are
> short-sighted.
>
> The reasoning for this assumption is:
>
> * Deployment of new mining hardware controlled by others may occur at any
> time you do not control.
>   * Thus, any transactions you leave on the table are potentially taken by
> somebody else and not by you.
>   * Sudden changes in hashpower distribution may reduce your expected
> future earnings, so any future theoretical earnings should be discounted
> (*in addition to* expected return-on-investment on getting money you can
> invest *now*).
>

Our analysis assumes constant difficulty, i.e., no significant changes of
the miners set. Indeed, hash-rate changes typically occur at a much larger
granularity than your average HTLC timeout. For instance, we noticed plenty
of lightning nodes use timeouts of a day. So, we do not consider
optimization at infinity, just a day ahead, and within this time frame all
the factors you mentioned are not expected to dramatically change.

That being said, it would be interesting to analyze the effect of miners
joining during the HTLC duration. Intuitively, this shouldn’t affect the
results, as those new miners have the same incentive to wait for the
higher-paying tx.


>
> It also strikes me that, in a world with RBF and CPFP, the same endpoint
> (i.e. miners earn the entire fund of the HTLC) is achieved by existing
> HTLCs, without the additional branch and script opcodes needed by MAD-HTLC.
> For example, if an HTLC is confirmed but the hashlock-claiming transaction
> is not being confirmed (because miners are holding it up because Bob is
> offering a much higher fee in the future for the timelock-claiming
> transaction), then Alice can, regardless of the reason why it is not being
> confirmed, bump up the fee with RBF or CPFP.
>
> If the fee bump offered by Alice is sufficiently large, then miners will
> start re-preferring the Alice hashlock transaction.
> To counter this, Bob has to bid up its version higher.
>
> As the timeout approaches, Alice can bump up its fee until it is just 1
> satoshi short of the total fund.
> It is rational for Alice to do so since at timeout, it can expect to lose
> the entire fund.
> In order for Bob to win, it has to beat that fee, at which point it equals
> or exceeds the total fund, and miners get the total fund (or more).
>
> Knowing this end-point, rational Bob will not even begin this game.
>
> I think this research considers these two endpoints to be distinct:
>
> * Bob misbehaves and the entire fund is punished by miners, leaving miners
> with the fund and Alice and Bob without money (MAD-HTLC).
> * Bob misbehaves, Alice counters, and the ensuing fee war leads to fees
> approaching the fund value, leaving miners with the fund and Alice and Bob
> without money (standard HTLC).
>
> But in practice I think both endpoints are essentially equivalent.
>

These are not the same scenario, since in HTLC there is a race between
Alice and Bob. Alice might not wish to pay the full HTLC amount once she
sees Bob is trying to cheat. She could wait until close to the timeout so
as to reduce the time Bob can respond. Of course Bob would do the same. So
this is an actual race, and Bob takes no risk since his payment is all from
the HTLC amount. Mutual destruction is only assured under certain
assumptions in HTLC. MAD-HTLC achieves security without relying on such
assumptions.


>
> --
>
> What MAD-HTLC can do would be to make different claims:
>
> * Inputs:
>   * Bob 1 BTC - HTLC amount
>   * Bob 1 BTC - Bob fidelity bond
>
> * Cases:
>   * Alice reveals hashlock at any time:
> * 1 BTC goes to Alice
> * 1 BTC goes to Bob (fidelity bond refund)
>   * Bob reveals bob-hashlock after time L:
> * 2 BTC goes to Bob (HTLC refund + fidelity bond refund)
>   * Bob cheated, anybody reveals both hashlock and bob-hashlock:
> * 2 BTC goes to miner
>
> This is an actual improvement over HTLC: Bob misbehavior leads to loss of
> the fidelity bond.
> The above cases can be assured by requiring both Alice and Bob to sign in
> the alice-hashlock branch, so that the splitting of the fund is enforced,
> and SegWit signing so that the dependent transaction is signed before the
> HTLC-funding transaction is.
> It can also be implemented with `OP_CHECKTEMPLATEVERIFY`.


The cases you present are exactly how MAD-HTLC works. It comprises two
contracts (UTXOs):
* Deposit (holding the intended HTLC tokens), with three redeem paths:
- Alice (signature), with preimage "A", no timeout
- Bob (signature), with preimage "B", timeout T
- Any entity (miner), with both preimages "A" and "B", no timeout
* Collateral (the fidelity bond, doesn't have to be of the same amount)
- Bob (signature), no preimage, timeout T
- Any entity 

Re: [bitcoin-dev] MAD-HTLC

2020-06-23 Thread ZmnSCPxj via bitcoin-dev
Good morning Itay, Ittay, and Matan,

I believe an unstated assumption in Bitcoin is that miners are short-sighted.

The reasoning for this assumption is:

* Deployment of new mining hardware controlled by others may occur at any time 
you do not control.
  * Thus, any transactions you leave on the table are potentially taken by 
somebody else and not by you.
  * Sudden changes in hashpower distribution may reduce your expected future 
earnings, so any future theoretical earnings should be discounted (*in addition 
to* expected return-on-investment on getting money you can invest *now*).

It also strikes me that, in a world with RBF and CPFP, the same endpoint (i.e. 
miners earn the entire fund of the HTLC) is achieved by existing HTLCs, without 
the additional branch and script opcodes needed by MAD-HTLC.
For example, if an HTLC is confirmed but the hashlock-claiming transaction is 
not being confirmed (because miners are holding it up because Bob is offering a 
much higher fee in the future for the timelock-claiming transaction), then 
Alice can, regardless of the reason why it is not being confirmed, bump up the 
fee with RBF or CPFP.

If the fee bump offered by Alice is sufficiently large, then miners will start 
re-preferring the Alice hashlock transaction.
To counter this, Bob has to bid up its version higher.

As the timeout approaches, Alice can bump up its fee until it is just 1 satoshi 
short of the total fund.
It is rational for Alice to do so since at timeout, it can expect to lose the 
entire fund.
In order for Bob to win, it has to beat that fee, at which point it equals or 
exceeds the total fund, and miners get the total fund (or more).

Knowing this end-point, rational Bob will not even begin this game.

I think this research considers these two endpoints to be distinct:

* Bob misbehaves and the entire fund is punished by miners, leaving miners with 
the fund and Alice and Bob without money (MAD-HTLC).
* Bob misbehaves, Alice counters, and the ensuing fee war leads to fees 
approaching the fund value, leaving miners with the fund and Alice and Bob 
without money (standard HTLC).

But in practice I think both endpoints are essentially equivalent.

--

What MAD-HTLC can do would be to make different claims:

* Inputs:
  * Bob 1 BTC - HTLC amount
  * Bob 1 BTC - Bob fidelity bond

* Cases:
  * Alice reveals hashlock at any time:
* 1 BTC goes to Alice
* 1 BTC goes to Bob (fidelity bond refund)
  * Bob reveals bob-hashlock after time L:
* 2 BTC goes to Bob (HTLC refund + fidelity bond refund)
  * Bob cheated, anybody reveals both hashlock and bob-hashlock:
* 2 BTC goes to miner

This is an actual improvement over HTLC: Bob misbehavior leads to loss of the 
fidelity bond.
The above cases can be assured by requiring both Alice and Bob to sign in the 
alice-hashlock branch, so that the splitting of the fund is enforced, and 
SegWit signing so that the dependent transaction is signed before the 
HTLC-funding transaction is.
It can also be implemented with `OP_CHECKTEMPLATEVERIFY`.

Regards,
ZmnSCPxj
___
bitcoin-dev mailing list
bitcoin-dev@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev


[bitcoin-dev] MAD-HTLC

2020-06-23 Thread Stanga via bitcoin-dev
Hi all,

We'd like to bring to your attention our recent result concerning HTLC.
Here are the technical report and a short post outlining the main points:

* https://arxiv.org/abs/2006.12031
* https://ittayeyal.github.io/2020-06-22-mad-htlc

Essentially, we find that HTLC security relies on miners being altruistic,
or at least myopic. This might be enough for some time, but it took us 150
lines of code to make bitcoind non-myopic.

On the positive side, we discovered an alternative to HTLC that we call
MAD-HTLC, which is provably secure -- everyone's best interest is to behave
as desired.

We've notified relevant teams in advance.

We'll appreciate any comments.

Best,
Itay, Ittay, and Matan
___
bitcoin-dev mailing list
bitcoin-dev@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev