Exploring Locality in Ethereum Transactions

Abstract

Early open blockchain designs face low throughput, high latency, and prohibitive costs for setting up a full node. New designs improve this with innovative mechanisms for handling transactions and the blockchain state, often assuming locality properties in the workload of transactions. Temporal locality allows efficient space management such as light nodes or snapshot-based bootstrap. Disjoint access parallelism, which depends on spatial locality, enables parallel processing of non-conflicting transactions. We analyze locality properties and their interplay in the largest transactional workload available to date, that of Ethereum. Our results show that, although transactions generally display good locality, a minority of accounts are responsible for caching- or parallelism-unfriendliness, calling for specific identification and handling in future blockchain designs.