Estimating Bitcoind's Share of Blocks

bitcoind is a daemon process that handles sending and receiving bitcoins for Bitcoin Core (formerly the “Satoshi client”) and its various forks (XT, Classic, Unlimited). Some people who use bitcoind to manage their funds use Bitcoin-Qt (sometimes also referred to as “Bitcoin Core”), a graphical interface for bitcoind available on many popular operating systems.

Based on a sample of 150k transactions on July 2nd and July 6th of this year, I’d estimate that between 20% and 26% of active Bitcoin wallets are using bitcoind.

This data is based on the distinct fingerprint currently generated by bitcoind when drafting new wallet transactions, as well as clustering data from WalletExplorer.com.

My colleague Antoine Le Calvez [github | twitter | p2sh.info] has used similar fingerprinting data to graph the prevalence of bitcoind transactions over time between December 2nd, 2014 and October 17th, 2016:

Note that the fingerprint characteristics that Antoine sampled in the above chart only became effective after the release of Bitcoin Core 0.11.0 on July 10, 2015 around block height 364700, right when the plot line starts to take off.

I determined the sample size based on what subjectively seemed like “enough” after running for several hours (bottleneck: clustering API). I picked the dates fairly at random, but tried to pick two fairly dissimilar days of the week and times to reduce time zone bias. A spreadsheet of data is available here: https://docs.google.com/spreadsheets/…

Using clustering data in this fashion is not a perfect estimation, and may lead to the under-counting or over-counting of bitcoind wallets. This estimation does not necessarily reflect the number of bitcoind users, as it includes both corporate services and end users, both of which may actively use multiple wallets.

Full source code and instructions for running the tool that generated data is available here: https://github.com/kristovatlas/estimate-fullnode-users