Isaac Sheff

Senior Research Scientist

[Isaac Sheff @IsaacSheff]

@getsqt For A validators to agree with each other (with A's integrity), all A validators participate. The atomicity guarantee is that A validators will agree with B validators. The more validators A & B have in common, the stronger that guarantee. 2/2

[Isaac Sheff @IsaacSheff]

New blog post about my main research project, Chimera Chains!

Chimera Chains generalize shared sequencers. They use a consensus based on Heterogeneous Paxos to simplify cross-blockchain atomic transactions for chains with overlapping trust models. 1/9

https://t.co/M2gkOy3amg

[Isaac Sheff @IsaacSheff]

@getsqt In brief: suppose we have 2 base chains, A & B. All validators want to agree on the chimera state (in part because it can eventually effect base partition state). For B validators to agree with each other (with B's integrity), all B validators participate. 1/2

[Isaac Sheff @IsaacSheff]

@getsqt It depends what you mean by "validate" here.

For the chimera chain, the relevant thing about overlapping participants is that the overlap determines the atomicity guarantee. To maintain a replicated state machine, everyone still has to maintain and update a replica of the state

[Isaac Sheff @IsaacSheff]

@getsqt Part of starting up a chimera chain is getting all the participants to participate. This means participating in the chimera mempool, chimera heterogeneous consensus, and chimera replicated state machine. Ultimately, the Typhon software seeks to make this easy.

[Isaac Sheff @IsaacSheff]

@zkmattwyatt @anoma I'm not familiar with electricity market contracts, but it does sound like you're describing a cross-database atomic transaction. These are usually solved with multi-phase commit, which we talk about in the blog post.

https://t.co/I0QsD4Y4T6

[Isaac Sheff @IsaacSheff]

@getsqt Actually all the consensus participants in both chains participate, but yes.

[Isaac Sheff @IsaacSheff]

@getsqt Chimera chains use Heterogeneous Paxos, which sets the leader selection protocol (for the chimera chain) for all participants.

[Isaac Sheff @IsaacSheff]

@ArafatShymaa In a setup with chimera chains, transactions do form a partially ordered set, but I wouldn't say "Paxos is similar to Posets." Paxos, or in this case Heterogeneous Paxos, is a consensus algorithm used to produce a total order within each (chimera) chain. https://t.co/YJfFaUxp6z

[Isaac Sheff @IsaacSheff]

@nitanshu Your intuition is dead-on. A core problem here is avoiding race conditions between different ordering mechanisms / chimera chains. Our "partitions" can be seen as "lock holders," ensuring at most one ordering mechanism updates any given piece of state at any given time.

[Isaac Sheff @IsaacSheff]

@ArafatShymaa thanks for the bug report! I'll pass this along.

[Isaac Sheff @IsaacSheff]

@ArafatShymaa I suppose it depends what you mean by "verify," but yes, the core difficulty is avoiding / resolving conflicts from cross-chain transactions. Multiple L2s / rollups can face the same difficulty (although in their case their shared L1 may be able to help resolve such conflicts).

[Isaac Sheff @IsaacSheff]

@zkmattwyatt @anoma I'm not sure about the router analogy. Chimera chains are for atomic cross-chain transactions. "Traditional" analogues are atomic cross-database transactions, which are also hard. Asynchronous messaging between chains feels a lot more like routing, and can be done with IBC. 3/3

[Isaac Sheff @IsaacSheff]

@zkmattwyatt @anoma Transactions committed to the chimera chain are final. They do not then have to "make it to the base chain." However, chimera chains have a different notion of "committed" for each base chain. The atomicity guarantee is the guarantee that the base chains' "commits" agree. 2/3

[Isaac Sheff @IsaacSheff]

@bgmshana Thanks!
Chimera chains in particular are the collaboratively produced blocks. We call original chains (which define the trust models) "base chains." So when base chains' validators overlap sufficiently, they can form a chimera chain, and get cross-domain atomicity guarantees.

[Isaac Sheff @IsaacSheff]

@CBarraford @CannnGurel @iceman00008 The chimera partitions are managed by a heterogeneous paxos where, it's true, a proposer chooses which TXs are proposed. The same mitigations (e.g. Narwhal) apply as in other chains. I'm not sure what edge cases you're referring to, or what you mean by "spray and pray."

[Isaac Sheff @IsaacSheff]

Chimera chains are intended to "make the easy thing easy." When a set of chains have a lot of overlap, it should be fast and easy to do cross-chain transactions, and with Chimera chains, it will be. Any questions? Come discuss on the forum. 9/9

https://t.co/MiugxFhfBc

[Isaac Sheff @IsaacSheff]

In the blog post, we detail what it means to keep or lose atomicity, what atomicity guarantees we can expect will be available in the wild, and future generalizations of Chimera Chains. We also discuss communication within atomic bundles, which shared sequencers can use too. 8/9