Ethereum execution layer researcher weiihann has released a performance experiment report on state expiry. According to Foresight News, the experiment involved running approximately one year of Ethereum mainnet block load on the Geth client, comparing the performance differences between 'full state nodes' and 'nodes retaining only one year of active state.' The study found that reducing state size not only lowers hardware requirements but also provides room for increasing gas limits and throughput.
The experiment results indicate that retaining only active states accessed within one year reduces the database size from 359 GB to 81 GB, a decrease of about 77.5%, with the most significant reduction observed in storing Trie nodes. In terms of performance, block re-execution time was shortened by approximately 15%, and read performance improved significantly, particularly with storage read P50 latency reduced by 46% and P99 latency by 36%. Additionally, tail latency improved, with P99 block insertion time reduced by 21%, aiding nodes in maintaining synchronization under load. Future research will explore comparisons with other clients, different expiry cycles such as six months, and strategies focused solely on contract storage cleanup.