dataforge-labs/bitcoin-network-propagation

Bitcoin network propagation

How fast the Bitcoin peer network learns things, measured from connections held to roughly 120 peers at once.

Nothing here can be reconstructed later. A block carries only the timestamp its miner claimed, an announcement carries none at all, and a peer's own relay policy is broadcast and forgotten.

Contents

name	one row is
e21_btc_block_propagation	one peer announcing one block, timestamped
e21_btc_tx_propagation	one peer announcing one sampled transaction, timestamped
e21_btc_relay_floor	one peer's own minimum relay fee, at the moment it announced it
e21_btc_p2p_peers	one peer connected to: user agent, services, handshake state

Blocks against transactions

These propagate on completely different timescales, and the contrast is the point.

Across four blocks in one window, half the peers had a new block within about 0.4 seconds, with a tail out to 52 seconds on one peer. Transactions took a median of about 23 seconds between the first and last peer, p90 71 seconds, and a maximum of 103.

That gap is not congestion. Nodes deliberately randomise transaction announcement timing so that a watching node cannot work out where a transaction entered the network. Blocks get no such treatment, because there is nothing to hide. So the transaction table measures a privacy mechanism and the block table measures raw relay speed.

Sampling and floors

Transactions are sampled by txid, roughly one in sixty-four, because announcements otherwise arrive in the thousands per second. The test is a property of the hash, so it is the same on every peer and a curve is never truncated by when we started watching. Only the announcement is recorded; the body is never requested and is on chain anyway once mined.

e21_btc_relay_floor is the other half of whether a transaction can move at all. Below a peer's minimum relay fee it is not forwarded, so it never reaches a miner. One sample of 41 peers held 8 distinct floors, with only 34% at the common 1.0 sat/vB default, 56% below it and 10% above. The non-round values are nodes whose own mempool is evicting.

Before you build on this

• Timings are ours and include network distance to each peer, which varies with geography. Differences of milliseconds are partly distance; differences of seconds are not. peer_addr is retained so this can be controlled for.
• Roughly 120 peers out of tens of thousands reachable. A sample of the network, not the network.
• A peer that disconnects stops announcing, which resembles slowness. e21_btc_p2p_peers carries handshake state so a gap can be told from a silence.
• Relay floors flatten while the mempool is quiet, because nodes then sit at whatever they were configured with. They move when a mempool fills, which is when they matter.

Partitions are parquet, one file per collection window, under dataset/YYYY/MM/. Every dataset here carries a FIXED 7-day sample WINDOW starting at its own first day of collection, together spanning 2026-08-29 to 2026-09-04, so you can check schema, coverage and quality before asking for more. It does not advance, so there is nothing to gain by re-downloading it. The full history is held privately, available on request.

from huggingface_hub import snapshot_download
import pandas as pd, glob

path = snapshot_download("dataforge-labs/bitcoin-network-propagation", repo_type="dataset",
                         allow_patterns="e21_btc_block_propagation/**")
df = pd.concat(map(pd.read_parquet,
                   glob.glob(f"{path}/e21_btc_block_propagation/**/*.parquet", recursive=True)))

Coverage

e0_run_manifest lists every collection window with its poll counts and failure counts, and is published in full rather than windowed. Gaps between windows are real, cannot be filled in afterwards, and nothing here is interpolated.

License and contact

ODC-BY: use it freely, credit "DataForge (dataforge-labs)". Questions and requests for the full history via the discussions tab.