Mar 24, 2023

Some initial log-to-graph results... I completed the implementation of tx hash computation from the message buffer, which doesn't appear to be a material gain. Also discovered I wasn't requesting witness blocks, so updated that. This plot shows a few metrics at 10k block intervals up to 300k. Time and tx count are scaled for presentation. It appears that in this range at least block download, parse, hash and archive (time) is linear relative to the amount of data (archive [total store], input, output, put [input + output] sizes), and objects (tx count) in the store. Each of these lines very closely overlays the other with a constant scale factor. I'm doing a full chain run now to verify that this holds for the entire chain. This is all still just a single peer connection with sequential block download (against a pretty consistent peer).