Data-driven analysis on EIP-7907

Execution Layer Research
5

  1. For this comment, I actually meant Pebble’s internal cache, not Geth’s cache. You can tune Pebble’s cache size directly and make it smaller, for example:

    Cache: pebble.NewCache(int64(cache * 1024 * 1024))

  2. Regarding your interpretation:

    This is possibly correct. I’ve observed the same phenomenon with RocksDB as well, although I don’t fully understand the underlying mechanism yet :slight_smile:

  3. From my experience, cache size is the dominant factor affecting DB performance. See our latest research on how to tune the cache size Demystifying Blockchain KV Lookups: From O(log N) to O(1) Disk I/O . So my suggestion is:

    1. Explicitly tune (reduce) Pebble’s cache size
    2. Run another program that consumes most of the system memory, to ensure the effective cache available to the DB benchmark remains small. FYI, I’ve previously built a small Rust tool for this purpose: revm/bins/dbtool/src/cmd/cache_killer.rs at 3ec984266af98838967a88c2079bac2cebd7c48d · dajuguan/revm · GitHub