Abstract
The Bitcoin peer-to-peer (P2P) network relies on a decentralized topology that intentionally obscures its structure to prevent adversarial attacks. While reachable nodes are easily indexed, "unreachable" nodes—those behind firewalls or NATs—remain largely invisible to standard crawlers. This paper introduces the Passive Announcement Listening (PAL) method, which leverages the network's internal gossip protocol to estimate the population of these hidden peers. By monitoring unsolicited address announcements over a five-year period (2015–2020), we find that unreachable peers providing "useful services" numbered approximately 31,000 daily by late 2020, outnumbering reachable peers by a factor of six.
Introduction
The Bitcoin P2P network consists of two primary node types: reachable peers, which accept incoming connections, and unreachable peers, which can only initiate outbound connections. Unreachable peers are vital to the network's health, contributing to block and transaction propagation, yet their "invisible" nature makes them difficult to quantify. Accurate network models are essential for simulating protocol optimizations and assessing resilience against eclipse attacks. Previous estimates of unreachable peers have lacked consistency, ranging from 16,000 to 155,000 nodes depending on methodology.
Methodology: Passive Announcement Listening (PAL)
The PAL method shifts from active crawling to passive observation. A monitor node is established to maintain connections with all known reachable peers. Instead of probing for new addresses, the monitor listens for addr (address) messages propagated through the network's gossip protocol.
The core logic of PAL relies on filtering for "small" addr messages containing 10 or fewer entries. In the Bitcoin protocol, these small messages typically represent unsolicited self-announcements or recently received announcements from active peers. By aggregating unique IP addresses from these messages over 24-hour intervals and subtracting known reachable peers, the population of active unreachable peers can be estimated.
Validation and Data Analysis
The study utilized a longitudinal dataset spanning 2015 to 2020. To validate the PAL method, the researchers operated a "public peer" (probe) for 16 months, directly observing incoming connections from unreachable peers. This ground-truth data revealed that PAL detects approximately 50% of unreachable peers providing "useful services" (e.g., NODE_NETWORK). This detection rate is attributed to the random 24-hour mean interval for announcements and the protocol's requirement that forwarded announcements be "fresh" (less than 10 minutes old).
Key Results
- Population Estimate: By the end of 2020, the network hosted approximately 31,000 unreachable peers providing useful services daily.
- Historical Trends: Unreachable peer counts peaked at roughly 72,000 during the late 2017 Bitcoin price surge before stabilizing between 31,000 and 37,000 in 2020.
- Detection Efficiency: The PAL method is highly reliable for reachable peers, detecting 98% of them, and captures roughly 50% of "useful" unreachable peers.
- Network Anomalies: Significant spikes in announcements in late 2018 suggest periods of "bogus address" flooding, highlighting the network's susceptibility to gossip poisoning.
Discussion and Limitations
The PAL method is specifically tuned to detect peers providing "useful services," as Bitcoin Core nodes generally do not forward addresses of peers lacking these flags. Consequently, the total number of unreachable peers (including those without full blockchain storage) is likely higher. Furthermore, the reliance on unauthenticated addr messages means the method can be influenced by attackers injecting fake addresses into the gossip stream.
Future research will focus on refining the 50% detection constant through deeper simulations of addr propagation, aiming to provide a more precise scaling factor for total network size estimation.
Summary updated