EVM Workloads in the Wild: Evidence for Multi-Dimensional Gas Metering, State Growth, Delayed Execution, and Parallelism
Lioba Heimbach
Session summary
The presenter reports a large-scale measurement study of real EVM workloads on Ethereum mainnet and the Base L2 during 2025, plus a focused September 2025 dataset. They analyze opcode-level gas decomposition, state growth (accounts, storage slots, contract code), and transaction categories to test the EVM’s implicit assumption that execution conditions and resource mixes stay stable. The study compares Ethereum and Base, and tracks how Ethereum’s workload changes over the year, including around a capacity increase.
According to the measurements, Ethereum and Base share the same opcode pricing but exhibit different and evolving resource profiles. Ethereum L1 appears more storage‑write heavy, while Base uses larger fractions of storage reads and compute and more gas per transaction. On Ethereum, storage writes and reads decline as compute’s share rises, and a capacity doubling is followed by non‑proportional growth across resource dimensions and over‑proportional state growth. The September simulations show that gas use and access patterns vary across nearby states for a notable share of transactions, especially in DeFi and MEV, which the presenter frames as a UX concern for users relying on pre‑trade simulations.
Key takeaways
- Measurements suggest that a single EVM gas schedule applied across chains does not match observed resource mixes, which differ between Ethereum and Base and drift over time on Ethereum.
- Following Ethereum’s 2025 capacity increase, resource usage and persistent state growth did not scale uniformly, with storage growth rising more than proportionally to gas target changes on both chains.
- Simulating September 2025 transactions on older states reveals substantial state sensitivity for DeFi and MEV, raising concerns for designs and users that assume stable gas estimates and execution paths.