Dissecting the EIP-2930 Optional Access Lists

Lioba Heimbach, Quentin Kniep, Yann Vonlanthen, Roger Wattenhofer, Patrick Züst

Abstract

Ethereum introduced Transaction Access Lists (TALs) in 2020 to optimize gas costs during transaction execution. In this work, we present a comprehensive analysis of TALs in Ethereum, focusing on adoption, quality, and gas savings. Analyzing a full month of mainnet data with 31,954,474 transactions, we found that only 1.46% of transactions included a TAL, even though 42.6% of transactions would have benefited from it. On average, access lists can save around 0.29% of gas costs, equivalent to approximately 3,450 ETH (roughly US$ 5 Mio) per year. However, 19.6% of TALs included by transactions contained imperfections, causing almost 11.8% of transactions to pay more gas with TAL than without. We find that these inaccuracies are caused by the unknown state at the time of the TAL computation as well as imperfect TAL computations provided by all major Ethereum clients. We thus compare the gas savings when calculating the TAL at the beginning of the block vs. calculating it on the correct state, to find that the unknown state is a major source of TAL inaccuracies. Finally, we implement an ideal TAL computation for the Erigon client to highlight the cost of these flawed implementations.

Introduction

In 2020, Ethereum implemented EIP-2929, which significantly increased the gas costs for "cold" state accesses—accessing an address or storage key for the first time in a transaction. To mitigate the impact on complex smart contracts, EIP-2930 introduced Transaction Access Lists (TALs).

A TAL allows a user to pre-declare which addresses and storage keys their transaction will touch. In exchange for paying a small upfront fee for the list, the actual execution costs for those items are reduced (treating them as "warm" accesses). Beyond immediate gas savings, TALs are a prerequisite for parallelization; if the network knows exactly what a transaction will touch beforehand, it can safely run non-overlapping transactions at the same time. This study investigates why this feature hasn't seen widespread adoption and evaluates the effectiveness of current TAL implementations.

Methodology

The research follows a multi-faceted approach:

  1. Longitudinal Analysis: Tracking TAL usage on the Ethereum mainnet from April 2021 through August 2023.
  2. Deep Dive & Simulation: Using a full month of mainnet data from August 2023 (~32 million transactions), every transaction was re-executed at least six times using different TAL configurations (empty, original, and client-generated) to measure potential vs. actual gas savings.
  3. Client Auditing: Analyzing the source code of major Ethereum clients (Geth, Nethermind, Besu, and Erigon) to evaluate TAL generation logic.
  4. Optimization: Implementing an "ideal" TAL generator within the Erigon client to establish a benchmark for maximum possible efficiency.

Despite the potential benefits, adoption remains remarkably low. While there has been a gradual increase in TAL usage over time, the overall percentage remains a tiny fraction of total network activity.

TAL usage over time from April 2021 to July 2023.

Key Findings

The "Warm" Address Trap and Client Flaws

Ethereum automatically considers certain addresses "warm" (and thus cheap to access), such as the transaction sender, the recipient, and "precompiles." Many popular Ethereum clients mistakenly include these in the TAL. Since every entry in a TAL costs gas, including an address that is already "warm" results in a net loss for the user. All major clients (Geth, Nethermind, Besu, Erigon) were found to have suboptimal TAL generation logic.

The State Uncertainty Problem

Generating a "perfect" TAL requires knowing the exact state of the blockchain at the moment of execution. However, because other transactions might execute before yours in the same block, the "intra-block state" is unpredictable. For 71% of transactions that could benefit from a TAL, the list cannot be computed correctly based only on the previous block's state.

Economic Impact and Potential Savings

Only 1.46% of transactions included a TAL, despite 42.6% of all transactions being eligible for savings. Furthermore, 11.8% of transactions using a TAL actually paid more gas than if they had provided no list at all due to imperfections.

Comparison of cumulative gas and ETH saved across different TAL implementations.

As illustrated in the figure above, the potential for savings is significant. If every transaction used an ideal TAL, the network could save approximately 3,450 ETH (~$5 million USD) per year.

Conclusion

The primary barrier to TAL adoption is the unpredictability of the intra-block state. Unless a user can guarantee their transaction's position in a block (e.g., via MEV searchers), generating a perfect TAL is difficult. To improve adoption, smart contract developers could provide TAL templates, and Ethereum's mechanism design might need to increase gas discounts to better incentivize use, ultimately enabling parallel transaction execution.

Summary updated

Abstract

Ethereum introduced Transaction Access Lists (TALs) in 2020 to optimize gas costs during transaction execution. In this work, we present a comprehensive analysis of TALs in Ethereum, focusing on adoption, quality, and gas savings. Analyzing a full month of mainnet data with 31,954,474 transactions, we found that only 1.46% of transactions included a TAL, even though 42.6% of transactions would have benefited from it. On average, access lists can save around 0.29% of gas costs, equivalent to approximately 3,450 ETH (roughly US$ 5 Mio) per year. However, 19.6% of TALs included by transactions contained imperfections, causing almost 11.8% of transactions to pay more gas with TAL than without. We find that these inaccuracies are caused by the unknown state at the time of the TAL computation as well as imperfect TAL computations provided by all major Ethereum clients. We thus compare the gas savings when calculating the TAL at the beginning of the block vs. calculating it on the correct state, to find that the unknown state is a major source of TAL inaccuracies. Finally, we implement an ideal TAL computation for the Erigon client to highlight the cost of these flawed implementations.
BibTeX
@inproceedings{heimbach2024dissecting,
  author = {Lioba Heimbach and Quentin Kniep and Yann Vonlanthen and Roger Wattenhofer and Patrick Züst},
  title = {{Dissecting the EIP-2930 Optional Access Lists}},
  booktitle = {FC},
  year = {2024},
  doi = {10.1007/978-3-031-78676-1\_16},
}
Figure