Open daniellm opened 1 year ago
Summary of discussion done in discord -
This is more of a performance analysis, and since right now we are in the early stage of implementation, this is low priority/ enhancement.
Ignacio (jsign) has done a similar analysis in the past here. The difference is we would like to go one step further and generate files containing the access operations done to the verkle trie, so we can replay it at ease.
Gballet also did something similar in this PR
After talking with Ignacio he found this useful and said that this could be a great quick-"e2e" test. Also he said "For geth tests and optimizations, we've been replaying some chain history which clearly is heavier... but we re interested in testing the whole thing: a) The state migration using the overlay tree method (thus having dual tree access + extra load migrating state), 2) After the transition finishes, keep going with the VKT with the full state (which I think is what you're trying to focus). Of course as you imagine the setup for that is not light enough to be pushed to your gh repo... so I think that above is good."
Verkle trees behave different than Hexary Merkle trees in terms of performance (both CPU and RAM). We need to tune the performance to match our use cases, by choosing appropriate data structures and algorithms. For that, it would be helpful to use real-world samples of access to the Merkle tree, and translate them to access to the Verkle tree.
Todo: