| Liveness vs Safety |
Carnot favors safety, and when the chain cannot make progress, a new overlay needs to be formed until the protocol can make progress. |
Ethereum favors liveness in the case of failure, by using a hybrid consensus protocol (Gasper). Gasper is a combination of Casper the Friendly Finality Gadget (Casper-FFG) and the LMD-GHOST fork choice algorithm |
| Forking |
Carnot does not allow temporary forks or reorgs. Finality is reached after two consecutive (pipelined) rounds. |
The Gasper hybrid consensus protocol allows for forks and makes reorgs possible as it allows progress to be made in the absence of finality. |
| How many nodes attest each proposal? |
In Carnot, all nodes participate in every proposal. However, aggregation of their signatures happens only at the root committee level. At lower levels, their participation is just a vote, which will be processed by the parent node (however without aggregation of their signature). |
All nodes vote in a given epoch. An epoch is divided into 32 slots, and in each slot, a committee of nodes will be in charge of attesting the block of that slot. The signatures are aggregated by nodes called aggregators. |
| Responsive protocol |
Carnot is a responsive protocol. It means Carnot operates at the speed of the wire in a happy path. No block generation time is required. |
Ethereum lacks responsiveness and uses slot times even if the protocol achieves consensus before the slot time. |
| Fault Tolerance |
Carnot tolerates n/4 Byzantine nodes overall. We expect that this number can be brought close to (n-1)/3 (future work). Moreover, with a single committee, Carnot has (n-1)/3 fault tolerance. Fault tolerance reduces with the number of faulty |
Ethereum can tolerate (n-1)/3 Byzantine nodes. |
| PoS attacks |
Since Carnot avoids chain reorgs (forks), it is not possible for an adversary to force the protocol to switch to another fork in order to maximize the adversary’s stake in the network. |
Since chain reorg is possible in Gasper, fork choice algorithms can be exploited by adversaries to maximize their rewards, and stake and make the network more centralized. Possible chain reorgs gives a large attack vector on the PoS mechanism. (https://eprint.iacr.org/2022/289.pdf), https://eprint.iacr.org/2021/1413.pdf |
| Complexity |
Due to the absence of a fork choice algorithm, Carnot is simpler than Gasper. |
Gasper consensus is highly complex as it uses Casper-FFG and LMD-GHOST. Moreover, it's getting even more complex by adding patches to its PoS mechanism each time a new attack is discovered. |
| Elastic Scalability |
Generally, protocols operate in small network sizes i.e., a few nodes to a few hundred nodes (Tendermint, HotStuff, etc), or thousands of nodes (Gasper, Algorand). Carnot can scale from a few nodes to thousands of nodes. This is very important when sibling networks (sub-networks) bootstrap from the Nomos chain. Elastic scalability allows subnetworks to scale (or shrink) gracefully without any change in the core of the protocol. |
Gasper cannot shrink and scale gracefully. For example, the number of validators has to be at least SLOTS_PER_EPOCH * TARGET_COMMITTEE_SIZE (recommended minimum of 111). |
| Practical Privacy |
Carnot’s committee overlay structure can be leveraged to parallelize ZK proof generation as part of future work. In this way, users will enjoy privacy with a reasonable performance. It should be noted that not all proof generation can be parallelized. It depends on the protocol being used. For example, the Schnorr protocol for proving knowledge of a discrete logarithm and Bulletproof protocols support the parallel generation of proof. |
Gasper’s committee structure cannot be used without significant changes in the consensus algorithm. |
| Interchain communication |
Being responsive and low-cost of attestation verification, makes the interchain communication efficient and secure |
Interchain communication and verification can be very slow. |