Ethereum PK Miner

Trying to increase throughput as much as possible in a cpu ethereum PK miner

Architecture

We have 3 types of threads:

• UI
  • handles visualization of application throughputs, from tries to specific functions.
• Worker
  • generates random number and checks in the xorfilter
  • if true, sends to checker thread
• Checker
  • checks addresses that got sent for double-checking with a sqlite db containing all ethereum addresses

Only one thread is necessary for the UI and the Checker each.

Worker threads should be as many as possible, i.e., N-2 where N is the number of cores, though in my 28 core system doing only 8 worker threads leads to only around 20% less throughput but way less cpu usage.

Steps to run

Bigquery public ethereum data

First, we need to get all existing accounts with some ether balance above 0. For that, we can fetch google's public ethereum data using bigquery.

Pre-requisites:

• a google cloud account
• python3 + pandas + google cloud library
• google-cli to authenticate the current session

then, run the script:

python3 data/query.py

It will take some time to fetch all the data.

Preparing the xorfilter

The xorfilter will be used for the 1st check of address existence, to build it run

cargo run --release prepare

Options:
  -c, --csv-path <CSV_PATH>    Solution file to expand [default: ./data/accounts.csv]
  -f, --fuse <FUSE>            Which binary fuse to use, must be a value of 8, 16, 32 [default: 16]
      --fuse-path <FUSE_PATH>  Where to save the fuse, defaults to `./data/xorfilter{fuse}` [default: ]
  -h, --help                   Print help

Preparing the sqlite db

Pre-requisites:

• sqlite cli

For making the csv file into a useful sqlite db, we can import it directly using sqlite commands:

sqlite3 data/data.db
.mode csv
.import data/accounts.csv accounts
create index accounts_address on accounts(address)

Running

cargo run --release miner

Options:
  -t, --threads <THREADS>      How many worker threads should be spawned, if empty will use the num_cpus crate [default: 0]
  -f, --fuse <FUSE>            Which binary fuse to use, must be a value of 8, 16, 32 [default: 16]
      --fuse-path <FUSE_PATH>  Where the fuse is saved, if empty will read `./data/xorfilter{fuse}` [default: ]
  -h, --help                   Print help

Results

On my laptop's i7-14700HX, running on 26 worker threads:

On 8 worker threads: