Bitcoin full node and query server
Dec 16, 2024
View in context in the libbitcoin Slack #general mirror.
evoskuil01:57
This is the same on a log10 scale. That opens up the low end so you can see it. Also note that these are not all zeroized at start. The first block time varies due to header sync time. In any case the rate is very consistent despite thread count because at the tail end it's almost all big blocks. IOW, sync rate appears to scale at a fairly constant multiple per thread.
2 replies · evoskuil · last Dec 16, 2024
evoskuil03:28
This observation isn't correct. The non-log graph shows that all thread counts tested are equally performant once the sync becomes network limited, that's why the lines parallel each other at the end of the graph. This is more obvious on the 50k window plot as is controls for downloaded block size. We actually need more threads when the blocks are smaller, so that we can process more concurrently in an effort to keep up with the network.
evoskuil03:32
We could say that the observation is correct up until the point where the node becomes network limited, but that's not the differentiation shown at the end of the plots above.