TL;DR
As Ethereum’s "state" (the record of all accounts and contracts) grows indefinitely, it threatens the network's decentralization by making it harder for new nodes to join. This paper proposes an Object-Oriented Bayesian Network (OOBN) to model the transition to "Stateless Ethereum," predicting that while "witness" data packets will increase network traffic, the ecosystem is likely to remain healthy and functional.
What Problem Does This Paper Address?
Ethereum currently operates in a "stateful" manner: every full node must store the entire history and current state of the network (balances, smart contract code, etc.) to validate transactions. This state grows "ad infinitum," leading to high storage costs and long synchronization times. This creates a barrier to entry, potentially centralizing the network around a few well-resourced entities.
"Stateless Ethereum" aims to solve this by introducing witnesses—data packets containing Merkle proofs that allow a node to verify a block without needing the full state locally. However, witnesses add significant data overhead to every block. The authors tackle the critical question: Will the increased network traffic from these witnesses degrade the health and efficiency of the Ethereum ecosystem?
Methods
The authors employ Bayesian Networks (BNs), a probabilistic graphical modeling approach, to handle the uncertainty and complexity of the blockchain ecosystem. Specifically, they use an Object-Oriented Bayesian Network (OOBN), which allows them to break the system into four modular, hierarchical sub-models:
- Block Creation: Models factors like difficulty, gas limits, and transaction counts.
- Witness Creation: Quantifies the size and time required to generate witnesses.
- Ethereum Network: Models node distribution, bandwidth, and latency across five global regions.
- Block Propagation: The "endpoint" model that determines if nodes can keep up with the head of the chain.
The model was quantified using a hybrid approach: empirical data was used where available, and expert elicitation (consulting with domain experts) was used to fill gaps, such as estimating the impact of node location on peer-to-peer latency.
Datasets
- Ethereum Mainnet Data: 26,595 blocks mined between November 26–30, 2019, used to learn the structure and conditional probabilities of block and witness creation.
- Network Statistics: Data from
ethernodes.org(node types and distribution) andWondernetwork(global ping/latency data). - Propagation Data: Historical block propagation times sourced from
ethstats.io,ethstats.net, and academic studies (Silva et al., 2020).
Key Insights and Ideas
- Ecosystem Health as a Probability: Rather than a binary "will it work?", the model provides a "health" score. Under basic statelessness, the model predicts a 56% probability of a healthy ecosystem.
- The Cost of Statelessness: By comparing the model with and without witnesses, the authors found a relative 7% decrease in "health" (from 60% to 56%) when witnesses are introduced. This quantifies the trade-off between lower storage requirements and higher bandwidth usage.
- Sensitivity to Node Status: The model revealed that "Node Status" (whether a node is up-to-date, syncing, or offline) and the ability to "Keep up with the head of the chain" are the most influential factors on overall network health.
- Modular Future-Proofing: The OOBN structure allows researchers to swap out sub-models. For example, if a new witness compression technique (like Verkle tries) is implemented, only the "Witness Creation" module needs to be updated.
Key Results
- Baseline Health: The Ethereum ecosystem is expected to remain healthy following the introduction of Stateless Ethereum, with the "healthy" state being the most probable outcome (56%).
- Worst-Case Scenarios: In a "what-if" scenario involving very large witnesses and non-mining nodes, the probability of a node keeping up with the chain dropped from 65% to 54%—a 17% relative decline.
- Propagation Bottlenecks: The model confirmed that network latency and node bandwidth remain the primary constraints on how quickly the network can reach consensus in a stateless environment.
Limitations and Future Work
- Outdated Data: The empirical data predates EIP-1559, which significantly changed Ethereum’s fee market and block size variability. The authors acknowledge that post-EIP-1559 blocks are more variable and can be twice as large.
- Technological Evolution: The model does not account for Verkle tries, a newer proposal that would significantly reduce witness sizes compared to the Merkle-Patricia trees used in this study.
- Expert Bias: While necessary for missing data, expert elicitation introduces subjective bias into the probability tables.
- Future Directions: The authors suggest incorporating "Regenesis" (a proposal to periodically reset or prune state) into future iterations of the model to see how it interacts with statelessness.
Summary updated