- disable deref. then measure new IO and find new bottlenecks (likely they will change) - parallel commitment - seems possible reduce amount of computations: https://github.com/erigontech/erigon/pull/19899