# ignacio (jsign) - Update 13 In the last two weeks I resumed previous work on multiple improvements to the [go-verkle](https://github.com/gballet/go-verkle) repository. The work is composed of multiple PRs: - [PR: implement Copy-On-Write in leaf nodes](https://github.com/gballet/go-verkle/pull/314). [Created a while ago, but was refactored] - [PR: leverage multiple cores while flushing nodes](https://github.com/gballet/go-verkle/pull/320) - [PR: leverage multiple cores while doing Commit() on internal nodes](https://github.com/gballet/go-verkle/pull/321) - [PR: levrage `batchFr` in proof generation, and more memory-efficient `Serialize()` implementation](https://github.com/gballet/go-verkle/pull/322) - [PR: some cleanup](https://github.com/gballet/go-verkle/pull/319) Despite that all these changes have positive results in our go-verkle and go-ethereum benchmarks, we don't see a good extrapolation of results in the replay benchmark which looks suspicious. We're planning to benchmark each test separately to get better understanding. I've still [open other changes in go-ethereum](https://github.com/gballet/go-ethereum/pull/147) that should also improve the replay benchmark; but we're going step by step first starting with the go-verkle changes first. At some point in the future, I'd like to be able to run this "replay benchmark" locally so I can inspect a bit better why using VKTs there is so slow compared with MPT. In theory, we're pretty sure that's because of the different cryptography primitives but it still worth checking. Guillaume has some good ideas to "fake" some specific work to test the theory.