Key Observations and Insights

1. Impact of Architectural Choices on Performance

• DAG-based Consensus Architecture (Hetu)

  • combines CometBFT with a customized DAG-based Narwhal consensus implementation
  • Enables parallel transaction processing, potentially offering higher throughput than sequential models

• Cosmos+Ethermint Architecture (0G, Evmos, Kava)

  • Each Ethereum transaction is wrapped into a Cosmos transaction for consensus processing
  • This approach introduces additional overhead, reducing overall processing efficiency

• Cosmos+Beacon API+Geth/Reth Architecture (Bera)

  • Wraps an entire Ethereum block payload into a single Cosmos transaction
  • Significantly reduces consensus layer transaction load, performing better across all test categories

2. Sei's Unique Modifications

• Sei has extensively modified Cosmos, Tendermint, and Go-Ethereum. These deep changes make it fundamentally different from standard Cosmos chains like 0G, Evmos, and Kava. As such, Sei's performance cannot be directly compared to other chains in this analysis.

3. Block Production in Cosmos+Ethermint

• Ethermint-based chains produce blocks based on Ethereum transactions' gas limits rather than gas used
• This occurs because the Cosmos consensus layer cannot calculate gas used during block production
• To prevent misuse of inflated gas limits, a minimum gas usage ratio (typically 50%) is enforced
• Evmos' higher TPS in ERC20 and Uniswap tests is primarily due to larger block size configuration

4. Performance Gap Between Implementations

• 0G and Kava 0G and Kava share similar block size configurations, but 0G achieves better TPS

• Hetu parallel processing approach potentially offers higher throughput by processing transactions concurrently in a DAG structure

5.Design Advantages of Different Architectures

• Bera's block-level payload processing avoids transaction-by-transaction consensus overhead
• Hetu's DAG-based consensus allows for parallel transaction ordering and processing
• These architectural decisions can provide significant performance advantages in different scenarios

Conclusion

The performance differences highlight the impact of architectural and implementation choices:

• Bera excels due to its block-level payload processing approach, which reduces consensus overhead.
• 0G demonstrates the advantages of refining critical components like estimateGas to improve transaction throughput.
• Evmos achieves high ERC20 and Uniswap TPS through increased block size, while Kava lags due to older CometBFT and less efficient gas estimation.
• Sei’s extensive customizations set it apart from other chains, making direct comparisons to standard Cosmos-based architectures inappropriate.
• Hetu's DAG-based parallel processing offers a promising approach for high-throughput applications