https://diyhpl.us/wiki/transcripts/scalingbitcoin/tel-aviv-2019/atomic-multi-channel-updates/

https://twitter.com/kanzure/status/1230929981011660800

Matteo got sick and couldn't come. His coauthors couldn't come either. So a talk was pre-recorded and will be played now. Our work is about realizing atomic multi-channel updates with constant collateral in bitcoin.

We all know that blockchains have scalability issues. Public verifiability means we have to store each transaction on the blockchain. The transaction rate of bitcoin is far from being satisfactory. We can reach up to about 10 transactions/second in bitcoin. Contrasted to centralized systems like Visa, there's a difference of at least a few orders of magnitude. This is a severe issue that undermines the widespread deployment of blockchain technology.

Some work has been done to solve this problem. We can classify existing proposals into two categories. The on-chain approach aims to design better consensus protocols like sharding which allows parallel processing of transactions on different blockchains. On the other hand, we have off-chain approaches that use off-chain protocols like lightning network and raiden network. Do we really need to store each transaction on the blockchain to achieve public verifiability? Maybe users can exchange messages off-chain p2p and then resort to the blockchain only in the event of disputes. This is the basis for ideas like lightning network, raiden network, bolt, perun, liquidity network, etc. Payment channels and payment channel networks are the research field that I am going to concentrate on in this talk.

You're probably familiar with this, but I'll give you an overview of payment channels. Alice and Bob open a payment channel. Both of them have to sign any transaction that spends the money from out of this channel. The second step is to ask Bob to sign a refund transaction that transfers back the coins to Alice. This is required to protect Alice from a malicious Bob that refuses to use the money in the channel. At this point, Alice can verify the transaction and upload on the blockchain the commitment or funding transaction that creates the channels. From now on, Alice can pay Bob by creating a transaction and signing it that transfers the money within the channel by changing the distribution of coins between Alice and Bob. The mechanism allows for creation of new states and revocation of previous states. The reactive security model involves punishment of the counterparty that cheats the other user. The agrieved party will get all of the money from the cheater in the event of cheating, and this is pre-committed to and Alice and Bob don't participate if they don't have those reactive security transactions or justice transaction. This is how we can get many transactions off-chain without compromising on security.

Alice can't establish a new channel with each user she wants to transact with because this would be cumbersome and financially unstable because each new channel requires locking coins. To pay other users, the idea is to have a payment channel network where transactions and payments can be routed over a multi-hop network. All these payments along each hop in the route must happen atomically, meaning they all succeed or all they fail. This atomicity is achieved in lightning network. An intermediary has no reason to forward payments, so fees are added to the network this way all parties are incentivized to participate in the protocol. This completes a high-level description of the design of payment channel networks.

The fundamental question is how can we make sure that the transactions happen atomically. The key mechanism is a contract called a hashtime lock contract (HTLC) available in bitcoin. Also there is a timelock. HTLCs allow Alice to pay Bob if Bob shows some preimage x such that H(x) = y before the timelock runs out, by which the transaction can be redeemed.

HTLCs can be used to perform atomic multi-hop payments in a payment channel network. The idea is simple. Alice's payment to Bob is conditioned on Bob revealing the preimage of the hash and he can't do that right now, but what he can do is pay the 1 BTC to Carol which is also conditioned on the same hash preimage again. Now Carol has an incentive to reveal the preimage secret to Bob, and then Bob has an incentive to reveal it to Alice to get his fee. In other words, either both transactions succeed or none of them do. This gets the atomicity property we are looking for.

The timing conditions must be carefully selected. Carol might wait until the last moment of her HTLC timelock contract, and at this point Bob needs time to recover the secret x and upload his transaction to the blockchain as well. For this reason, the timing conditions are critical to the contract and the multi-hop network. The contracts must set timelocks sufficiently apart. In lightning network, the delta is one day per hop. This is a fundamental point that I want you to remember because it's one of the motivations for the work I am going to present.

This concludes my introduction to payment channel networks. We have shown in previous work that this is not secure in a privacy-preserving setting.See the paper "Security and privacy issues in PCNs" or "Anonymous multi-hop locks for blockchain scalability and interoperability" or "Concurrency and privacy in payment-channel networks". We have discovered what we call an attack where a malicious party is on the path between the sender and the receiver, and .... To solve these two problems, which come from the mechanism underlying the lightning network design which is the same condition used along the path for sender to receiver. We have proposed a different approach where HTLCs are replaced by a cryptographic condition which can achieve privacy and security. Our cryptographic approach has already been implemented by lightning network and others.

Today I would like to focus on two limitations of PCNs. This is regarding the collateral required in the network. So far we have only focused on payments between senders and receivers on a single path. Is there a possibility to perform payments across multiple paths or different paths? Also, for each user the user has to lock a certain amount of money that for a time depends on the user's position in the path between the sender and receiver. There is a delay in the timing conditions used in the HTLCs of 1 day in the lightning network for instance. This is quite a lot of time where users have to lock a significant amount of money.

So far we have only focused on payments that go along a path between sender and receiver. But there are other kinds of payments that might be useful. In the lightning network community, they have put forward the concept of an atomic multi-path payment where a single payment is split along multiple paths in case there exists no single path with enough capacity to perform the payment. Our goal is to go beyond paths, and support for atomic channel updates across arbitrary topologies. If we could do that, we could implement entirely new classes of payments. Think about crowdfunding where multiple people all pay to the same person. This uses a topology different from the single path considered so far. Also think about hybrid payments where Alice pays to Bob bitcoin and Bob pays to Carol in dollars or fiat currency and this doesn't show up on the blockchain but then there's another payment from Carol to Debbie in bitcoin. Here we would see two payments in different channels not connected on the blockchain, but in fact connected just in fiat currency. Or think about balancing and automatic applications where you want to synchronize payments in an arbitrary graph based topology. Surely you can come up with other applications that use other graph structures.

For the payment to succeed, each user has to lock a certain amount of coins in fact the amount that have to be transferred in the payment for a certain amount of time, and this time depends on the user's position on the path. If the user is at the beginning of the path, then the user has to lock the coins in a time that is linear in the size of the path between the sender and the receiver.

This gives rise to a griefing attack where the attacker can perform a denial-of-service attack because each user has to lock coins based on their position of the path, and their position in the path becomes the amplification factor. With a path of 5 hops, the first node has to lock coins for 5 days in the worst case.

Our goal in this project is to support constant collateral points. We want users to lock coins for a certain amount of time in a way that doesn't depend on the length of the path. This would minimize griefing attacks, and so on. The question is can we realize constant collateral payments in bitcoin or other cryptocurrencies?

The conclusion in the Sprites paper was no. What is the minimum extension of the bitcoin scripting language that enables constant collateral payments? Well, in this work, we show it is not required to extend the bitcoin script language. We show it is possible to implement constant collateral payments in bitcoin. Today I will introduce an optimized update of the AMCU protocol that we presented a few months ago.

Say we want to atomically update two channels one between Alice and Bob and one between Carol and David. Say 8 coins have to be transferred from Alice to Bob and 7 coins from Carol to David. These channels can be located anywhere in the payment channel network, and don't necessarily have to be contiguous.

The first phase is to setup between Alice and Bob. The first transaction signed by Alice and Bob splits the amount of coins into two channels, so that one can be used in ACMU channel update. Then phase 2 is a lock. The idea is to make sure they can recover their money in case something goes wrong in the protocol. They lock another transaction so that after some amount of time the money gets sent back to the channel as before from which Alice and Bob can continue. In phase 3, the transaction happens. Coins are transferred from the channel to Bob. The input to the transaction is a fresh channel, which doesn't exist yet. This seems a little bit weird, but this is actually the key to achieve atomicity. What we have to do in phase 4 is to synchronize all the multi-in multi-out transactions that synchronize all channel updates. The output channels are generic atomically, which means they either or transactions can be fired or can no longer can. If you follow the protocol carefully, you might notice we still have a problem to solve. The lock transaction and enable transaction are both possible after payment. This is a situation we would not like to be in. So therefore we introduce a phase 5 transaction that takes away the coins from the output channels constructed in phase 4. The disable transaction (phase 5) which is to be signed before, allows any party involved in the protocol to cancel the atomic multi-channel update, and send coins back to their respective owners.

The application scenarios we have in mind require to synchronize multiple channels, so these have to be revealed. But this is not necessarily true for the transacted values or amount of money transferred in each channel. It could be that owners of the channels want to keep this information private. Can we do that? Yes.

I am going to introduce a variant of ACMU.

The idea is to split the channel so that you could create a channel with a minimal number of coins. This is the channel we use to synchronize the channel updates. The lock transaction in phase 2 allow Alice and Bob to retrieve the coins after a certain timeout if something goes wrong in the protocol. The phase 3 transaction takes two inputs, both of which correspond to channels that have not yet been constructed, therefore the consume transaction cannot yet be redeemed. The key point is that the channel is populated locally by Alice and Bob and they decide how much money has to be transferred in this channel update and this information doesn't have to be delivered to the others. But what Alice and Bob do have to reveal in the phase 4 enable transaction is just this ... and this one doesn't reveal how much money is to be transferred in the multi-channel update. The multi-in multi-out transaction takes 2 inputs, one is the Alice-Bob channel and the other one is the Carol-David channel. All the parties have to sign. The enable transaction generates two-output channels. The enable transaction is the one that is giving us atomicity, as before. The disable transaction is used to solve ambiguous state between the locked transactions.

We have a formal security analysis in the universal composability UC framework. ACMU achieves atomicity: in particular if the coins at one channel are ready to be sent to the expected receiver, then all channels are ready to forward payments, otherwise coins remain available to their owners. ACMU does not achieve relationship anonymity - every user in the path collaborates with each other, but we can achieve value privacy. Also, AMCU provides accountability- it's possible to show a proof of misbehavior pointing to the party responsible for failure in the protocol.

We can reduce the collateral to a constant, and synchronize multiple channels atomically in a way that is backwards-compatible with bitcoin. This enables entirely new classes of off-chain applications like crowdfunding and channel rebalancing and we hope many more will come from bitcoin developers. The paper is available online, you can take a look.

https://eprint.iacr.org/2019/583.pdf

Sponsorship: These transcripts are sponsored by Blockchain Commons.

Disclaimer: These are unpaid transcriptions, performed in real-time and in-person during the actual source presentation. Due to personal time constraints they are usually not reviewed against the source material once published. Errors are possible. If the original author/speaker or anyone else finds errors of substance, please email me at kanzure@gmail.com for corrections or contribute online via github/git. I sometimes add annotations to the transcription text. These will always be denoted by a standard editor's note in parenthesis brackets ((like this)), or in a numbered footnote. I welcome feedback and discussion of these as well.

Tweet: Transcript: "Atomic Multi-Channel Updates with Constant Collateral in Bitcoin-Compatible Payment-Channel Networks" https://diyhpl.us/wiki/transcripts/stanford-blockchain-conference/2020/atomic-multi-channel-updates/ @matteo_maffei @CBRStanford #SBC20