TL;DR
This paper introduces Hybrid-Compact Block (HCB), a novel protocol designed to significantly speed up block propagation in the Ethereum network. By combining a secondary transaction pool for low-fee transactions and a machine learning-based prediction module for missing transactions, HCB achieves a matched-block probability of 90%, reducing block propagation time by over 50% compared to current methods.
What Problem Does This Paper Address?
Block propagation speed is a critical factor for the scalability and efficiency of blockchain networks. In Bitcoin, "compact blocks" were introduced to speed this up by replacing full transactions with their hashes, assuming most nodes already have these transactions. However, this approach falters in Ethereum. Ethereum's smaller transaction pools, network latency, and miners' strategic behaviors lead to a much higher probability that receiving nodes won't have the transactions needed to reconstruct a compact block. This results in costly retransmissions, negating the benefits of compact blocks and potentially degrading performance. The paper aims to solve this by making compact blocks effective in Ethereum's unique environment.
Methods
The authors propose a Hybrid-Compact Block (HCB) protocol with two key innovations:
- Secondary Pool: Instead of simply enlarging the primary transaction pool (which can increase empty block rates and reduce TPS), HCB introduces a Secondary Pool. This pool stores low-fee transactions that are evicted from the primary pool. These transactions are still available for reconstructing full blocks when an HCB block is received.
- Missing Transactions Prediction Module: This module uses a Naïve Bayes classifier, trained on empirical data from the Ethereum MainNet, to predict which transactions are likely to be missing at a receiving node. These predicted missing transactions are then "piggybacked" onto the HCB block, reducing the need for separate retransmissions. The model considers features like gas fees, transaction age, and rank ratio.
HCB is designed to be fully compatible with the existing Ethereum network architecture.
Datasets
The authors collected data from the Ethereum MainNet using specially configured probe nodes. This data was used for:
- Training the Missing Transactions Prediction Model: 1,000,000 transactions were collected, with 80% used for training.
- Evaluating the Prediction Model: The remaining 20% of the collected transactions were used as a test dataset.
- Evaluating the HCB Protocol: Experiments were conducted over the Ethereum MainNet using a prototype implementation, comparing HCB against the basic compact block (BCB), simple compact block with Secondary Pool (SCB), compact block with prediction module (PCB), and the standard Block-Hash Propagation (BHP) protocol.
Key Insights and Ideas
- Ethereum's Unique Challenges: The paper clearly demonstrates that the factors hindering compact blocks in Ethereum (small transaction pools, network latency, miner selfishness) are distinct from Bitcoin and require a tailored solution.
- Enlarging Transaction Pools is Not Enough: Simply increasing the size of the transaction pool can backfire by increasing the empty block rate, thus reducing overall transaction throughput (TPS).
- Predictive Power of Naïve Bayes: The Naïve Bayes classifier, despite its simplicity and assumption of feature independence, proves effective in predicting missing transactions with high precision in the Ethereum context.
- Synergy of Secondary Pool and Prediction: The combination of the Secondary Pool (for stale transactions) and the prediction module (for selfish/late transactions) is crucial for achieving a high matched-block probability.
Key Results
- Matched-Block Probability: HCB achieves a matched-block probability of approximately 90%, significantly outperforming BCB (12%), SCB (23%), and PCB (28%). This is comparable to Bitcoin's compact block probability (around 85%).
- Block Propagation Time: HCB reduces block propagation time by more than half compared to the current Ethereum protocol (BHP), especially for larger blocks.
- Prediction Precision: The Naïve Bayes prediction model achieves a precision of 0.951, indicating high accuracy in identifying missing transactions.
- Matched-Transaction Rate: HCB achieves a matched-transaction rate of around 0.99, meaning most transactions in an HCB block are readily available.
Limitations and Future Work
- Naïve Bayes Limitations: The authors acknowledge that the Naïve Bayes classifier's assumption of feature independence is not perfectly accurate and leads to a ~10% chance of retransmitting transactions.
- Future Work: The paper suggests exploring more sophisticated classifiers like Semi-Naïve Bayes, Bayesian Networks, Random Forests, and Support Vector Machines to further improve prediction accuracy and reduce retransmissions. They also note that HCB is compatible with Ethereum 2.0 and could be beneficial there.
Summary updated