Glossary · Blockchain development

P2P (Peer-to-Peer)

Published 23 May 2025 · Updated 21 Oct 2026 · 14 min read

P2P (peer-to-peer) is a distributed architecture in which participants share resources directly with each other. In Bitcoin, a full node checks transactions and blocks against shared rules. Nodes connect directly to each other and relay transactions and blocks hop by hop, without a central server. This differs from P2P trading, where people trade cryptocurrency with each other.

  • P2P nodes relay data directly between peers, hop by hop, rather than through one central server.
  • More full nodes add copies and resilience, not throughput, because each checks every transaction.
  • In October 2026, trackers count 25,897 reachable Bitcoin nodes and 7,777 Ethereum execution nodes.

Peer-to-peer versus client-server networking

A peer-to-peer network shares resources directly among participants. The client-server model separates requesters from providers, with a central server supplying resources. A hybrid keeps only discovery centralized, as Napster did with its index. In a blockchain P2P network such as Bitcoin, each full node independently checks transactions before relaying them and verifies every block itself.

Figure 1 · Simulated networks of 2,000 nodesTest P2P connection resilienceChoose links and a failure pattern, then move the outage slider. Compare the connected, isolated and offline shares against one server's clients.
Links per node
Failures
60% offline
Peer-to-peer, 100 sample nodes40 of the 40 nodes still online can reach each other.
Client and server, 99 clientsThe server is up in this picture. It fails in 60% of random draws, and then every client is cut off.
Offline60%Chosen at random
Still connected40%In the largest group that can reach each other
Online but cut off0%Up, but with no path to the main group

After 80% random outages, 96% of survivors stay connected with eight links. With ten links, 99% stay connected.

The page team simulated random networks of 2,000 nodes. These results do not measure a real network.

Alternate paths let traffic bypass failed participants. In the page team's simulation, eight links preserved connectivity among 96% of survivors after 80% random outages. Removing the most connected 80% first left 66% of survivors connected. These results do not remove shared dependencies. In October 2026, Bitcoin Core ships seven DNS seeds and Geth nine mainnet bootnodes. Hosting and routing also concentrate exposure. In 2017, dedicated hosting networks held 56% of Bitcoin nodes and 28% of Ethereum nodes.

File sharing can gain capacity when participants contribute resources. Blockchain replication works differently. Each full node checks every transaction and block against consensus rules, adding another verified copy rather than more throughput. The 2019 Erlay paper found relay bandwidth grew linearly with connection count. Nor does every participant serve others. Many only make outbound connections to peers. A blockchain light client downloads headers and relies on full nodes for other data.

From Napster to Bitcoin

Early peer-to-peer services separated finding content from transferring it. File sharing, voice calls and digital cash then exposed different trust problems. The Bitcoin white paper connects P2P networking with preventing a double spend, rather than simply distributing files.

  1. 1999

    Napster combines index and sharing

    Shawn Fanning's music service launched with a central song index and direct file transfers. It closed in July 2001 after A&M Records v. Napster.

  2. July 2001

    BitTorrent introduces file swarms

    Bram Cohen released BitTorrent on 2 July 2001. Peers exchange file pieces in swarms, finding each other through trackers, distributed hash tables or peer exchange.

  3. January 2002

    The Sybil attack gets named

    John Douceur described how one faulty participant can present many identities. Without a logically centralized authority, such attacks remain possible except under unrealistic assumptions.

  4. 2003

    Skype connects voice callers

    Skype launched with supernodes and ordinary participants. Microsoft moved supernodes into its data centers in 2012. The service was retired on 5 May 2025.

  5. October 2008

    Bitcoin proposes electronic cash

    On 31 October 2008, Satoshi Nakamoto announced electronic cash without a trusted third party. The white paper describes participants leaving and rejoining freely.

  6. October 2026

    Bitcoin Core expands connection limits

    Bitcoin Core 32.0, tagged on 21 October 2026, raises the default connection limit to 200. Versions 31.x and older keep the 125-connection default.

How a node finds its peers

Joining requires an initial contact, not a complete directory. A new P2P node learns addresses, tries connections and keeps useful peer contacts. Bitcoin uses relatively random selection. Ethereum's discovery organizes contacts through a structured table, so peer-to-peer networks need not share one topology.

Seeds provide first contacts

In October 2026, a new Bitcoin Core node queries all seven DNS seeds. If no addresses arrive within 60 seconds, it uses its built-in list of 2,057 fixed seeds. DNS answers lack authentication.

Address gossip builds memory

In October 2026, Bitcoin's addr and addrv2 messages carry up to 1,000 addresses each. Its address table buckets entries by address range, and new entries also by source range.

Ethereum organizes discovery

In October 2026, Ethereum execution discovery uses discv4 and discv5 with Kademlia-like tables. Signed node records, bootnode contacts and DNS lists support discovery. Each bucket holds 16 peer records.

Tor and I2P connections

Since Bitcoin Core 22.0 in 2021, onion connections require Tor v3, and I2P is supported. Its documentation warns that I2P-only operation may increase exposure to Sybil attacks.

How transactions and blocks spread

A node passes new data to its connected peers, which pass it onward. This process is called gossip. The Bitcoin white paper describes broadcasting transactions to all nodes, but no node connects directly to everyone. P2P relay reaches the wider network through successive hops. Transaction announcements, block delivery and consensus messages can follow different rules within the same peer-to-peer system.

Figure 2 · Flooding on random networksFollow a message between nodesSet network size, outbound links and assumed time per hop. The bars show reach, with milestones for half, 90% and 99%.
Network
Outbound links per node
200 ms
Share of nodes reached after each hopHops from the first node. Dark bars mark 90% of the network or more.
Half the network4 hopsAbout 800 ms at 200 ms per hop
90% of the network4 hopsAbout 800 ms
99% of the network5 hopsAbout 1 s. Everyone by hop 5

Eight links reach 90% in four hops and 99% in five across 25,000 nodes. Bitcoin Core's October 2026 defaults distinguish full-relay and block-only links.

This hop-count simulation comes from the page team. Real relay adds validation, bandwidth constraints and random delays.

Bitcoin announces transactions with inv. A peer requests them with getdata, then receives tx. Random announcement delays obscure the sender. Compact blocks, introduced in August 2016, send short IDs for transactions peers mostly already hold. In 2013, median block arrival took 6.5 seconds, with 5% still waiting after 40 seconds. In high-bandwidth mode, compact relay takes half a round trip when nothing is missing. Fetching missing transactions can raise that to 1.5 round trips. Bitcoin Core 27.0 made encrypted transport the default in April 2024.

In October 2026, Geth sends full transactions to about the square root of its peer count, eight of 50. Others receive hashes and fetch missing data. Ethereum's consensus layer uses libp2p and gossipsub meshes of about eight contacts per topic for blocks and attestations. Attesters vote about four seconds into each 12-second slot. In June 2024, 98% of beacon blocks arrived within four seconds. A Vienna observer measured a 1.77-second median and 3.75-second 95th percentile from slot start in December 2025.

Network size in October 2026

Public P2P counts depend on crawler methods and exclude nodes that reject incoming connections. The table separates Ethereum's execution and consensus layers. Counts, software shares and connection defaults reflect October 2026 snapshots, not a complete inventory of everyone participating.

NetworkNodes countedPeers per nodeSoftware mix
BitcoinNodes counted25,897 reachable nodes, btcnodes.io, 21 October 2026Peers per nodeCore 32.0 defaults to 200 total, eight full-relay and two block-only outbound, October 2026Software mixbitdis.org, 21 October 2026: Bitcoin Core 83.84%, Bitcoin Knots 15.93%
Ethereum execution layerNodes counted7,777 execution nodes, Ethernodes, 21 October 2026Peers per nodeOctober 2026 defaults: Geth 50, Nethermind 50, Besu 25, Erigon 64 peersSoftware mix21 October 2026, Ethernodes: Geth 40.5%, Nethermind 34.2%, Reth 13.6%, Besu 8.2%, Erigon 1.7%
Ethereum consensus layerNodes counted5,426 consensus nodes, Ethernodes, 21 October 2026Peers per nodeOctober 2026: Lighthouse targets 200 peers, Prysm allows 70Software mix21 October 2026, Ethernodes: Lighthouse 52.9%, Prysm 17.9%, Teku 9.5%, Nimbus 8.8%, Lodestar 3.7%
Canton NetworkNodes counted835 total validators and 62 Super Validator nodes, Canton Foundation, October 2026Peers per nodeValidators connect to a synchronizer for encrypted message orderingSoftware mixOctober 2026 Foundation figures distinguish validator roles rather than client software shares

Software concentration is a separate question from headcount. These snapshots show how unevenly clients are represented. Trackers also use different methods and report different totals. ProbeLab reports 8,157 Ethereum consensus nodes for 20 October 2026, versus Ethernodes' 5,426 for 21 October 2026. Neither figure should be treated as a count of individual users.

Attacks on peer-to-peer networks

A Sybil attack gives one operator many identities, making apparent diversity misleading. An eclipse attack controls all connections to a target node, presenting a false view of the chain. These related threats exploit different weaknesses in P2P networking. Address diversity does not guarantee operator diversity. With independent picks, complete outbound capture has probability f to the power of k, for attacker share f and k outbound connections.

Figure 3 · Eclipse attack oddsEstimate complete connection captureChoose a node's outbound connections and an attacker address share. Compare the sample connection mix with the calculated chance of capturing every outbound slot.
Outbound peers
50%
Typical outbound peers, A attacker, H honestChance that all 8 outbound peers belong to the attacker: 1 in 256.Each pick lands on an attacker address with a chance of 50%. The picture shows the typical mix, 4 of 8.
Every outbound peer is the attacker1 in 25650% to the power of 8
At least one honest peer99.6%One honest peer can relay the real chain
Typical honest peers4.0Out of 8 outbound connections
Spread over address groupsBitcoin Core buckets addresses by /16 range, or by AS with optional asmap, and opens one outbound link per range. Version 0.10.1 in 2015 enlarged these tables.
Feelers and test-before-evictShort feeler connections check that stored addresses still work, so old good peers are not pushed out by new fake ones (2017 and 2018).
Block-relay-only and anchorsTwo extra outbound links carry blocks only (2019). Since 2021 the node reconnects to them after a restart.
Caps per subnet in GethSince Geth 1.8.0 (February 2018), at most 2 addresses from one /24 per bucket and 10 in the whole table.

Eight full-relay and two block-only connections reflect Bitcoin Core's October 2026 configuration. The model counts outbound slots, not every possible connection.

The model assumes independent random picks. Real attacks first fill the address table, raising the attacker's share.

In August 2015, Heilman, Kendler, Zohar and Goldberg demonstrated Bitcoin eclipses with at least 85% success using 8,192 addresses across 32 distinct /24 blocks. Bitcoin Core 0.10.1 fixes raised the botnet requirement to 163,000 addresses for 50% success. Later defenses include feelers in 2017, block-only links in 2019 and restart anchors in 2021. In 2018, Marcus, Heilman and Goldberg eclipsed Geth in 49 of 50 tests using two hosts. Geth 1.8.0 added per-bucket and per-table address limits on 14 February 2018.

The 2017 Hijacking Bitcoin study found that a BGP hijack of fewer than 100 prefixes could isolate half of mining power. Erebus, published in 2020, uses millions of shadow addresses to partition targets without routing changes. In 2014, Biryukov, Khovratovich and Pustogarov demonstrated transaction deanonymization, linking transactions to originating IP addresses with a few machines. P2P code also faces denial-of-service attacks. Bitcoin Core's 2024 and 2025 disclosures include a blocktxn crash fixed in version 25.0. Geth published malicious-message advisories in 2024, 2025 and 2026.

Networks without global gossip

Not every peer-to-peer network distributes transaction content globally. In Canton, validators connect to a synchronizer. Its sequencer timestamps and orders encrypted messages, routing them only to recipients. Its mediator collects confirmations from stakeholders. The synchronizer never sees plaintext transaction content. The Canton Network explainer describes this selective model, where uninvolved validators need not know a transaction exists.

NetworkWho receives a transactionWho checks itWho stores it
BitcoinWho receives a transactionEvery reachable node through hop-by-hop relayWho checks itEvery full node independently verifies the transactionWho stores itEvery full node stores the verified result
EthereumWho receives a transactionEvery execution node through transaction relayWho checks itAll execution clients check transactions, validators vote on blocksWho stores itAll execution nodes store the result
CantonWho receives a transactionOnly validators involved in the transactionWho checks itOnly the involved stakeholder validators confirm itWho stores itInvolved validators store contracts relevant to their parties

This design trades universal transaction visibility for stakeholder-only disclosure and confirmation. It changes who must receive, verify and retain information. Canton still runs a decentralized synchronizer, with 62 Super Validator nodes reported by the Canton Foundation in October 2026. The comparison is therefore about information flow, not just connection counts.

P2P trading and other meanings

Peer-to-peer networking describes communication between machines. A P2P exchange connects people trading assets, often through escrow. P2P payments and lending use the same abbreviation for user-to-user services. Payment apps such as Venmo or Cash App do not need a blockchain network, so the label alone says little about architecture.

Marketplaces with platform escrow

In October 2026, Binance P2P reserves cryptocurrency in escrow until both sides complete the trade. Fiat moves through traders' payment providers. Building a P2P exchange platform means building this marketplace model, not network relay.

Non-custodial trading models

In October 2026, Bisq never holds user funds, and its nodes use Tor by default. Deposits use 2-of-2 multisig. Hodl Hodl uses multisig escrow addresses for individual deals.

Closures and compliance costs

LocalBitcoins closed trading in February 2023. Paxful closed in April 2023 and reopened in May 2023. In October 2025, it announced closure by 1 November 2025, citing misconduct and compliance costs.

Common trading scams

Binance's October 2026 scam list includes fake payment proof, third-party payments, support impersonation and triangle scams involving two scammers. It also names chargebacks after cryptocurrency has been released.

Planning a P2P network integration

An implementation plan needs to distinguish protocol behavior from application requirements. We can help frame the work around connectivity, relay behavior, security and transaction privacy. A peer-to-peer node deployment also needs a recovery path when initial contacts fail. These areas deserve separate decisions, whether the application relies on public gossip or Canton's stakeholder-only transaction flow.

PixelPlex case icon

We designed and implemented an incentive layer for FileStar's storage network, which uses the IPFS protocol. Our Echo wallet work includes mobile connections over Tor and a desktop wallet supporting a full node. We also build on Canton, where nodes connect through a synchronizer instead of global gossip.

  1. Deployment and peer connections

    Deployment requirements include node configuration and a clear distinction between inbound and outbound connections. Defaults depend on the selected client and version.
  2. Peer and relay monitoring

    Monitoring needs to distinguish transaction relay from block delivery. Track peer counts, inbound and outbound slots, and how fast transactions and blocks arrive.
  3. Eclipse and crash defenses

    A security plan covers address selection and relevant client fixes. Connection diversity and crash resistance address different failure modes.
  4. Origin privacy

    Bitcoin Core 31.0 (April 2026) added an opt-in mode that sends sendrawtransaction submissions only over Tor or I2P. Wallet transactions are not covered. Encrypted transport is a separate protection.
  5. Recovery from lost contacts

    A recovery plan accounts for failed seeds and providers. Fixed addresses and alternative contacts matter alongside hosting and routing dependencies.

Discuss the connectivity your P2P project needs

We can review your network model and help outline deployment, privacy and recovery requirements before implementation begins.

Discuss your project

P2P FAQ

What does P2P mean in cryptocurrency networks?

P2P means participants deal directly with each other. In networking, blockchain nodes connect to each other and relay transactions and blocks without a central server. In trading, P2P means users trade with each other, often through a platform's escrow.

How does P2P differ from client-server networking?

In client-server networking, clients send every request to a central server. If it fails, those clients are cut off. In P2P, nodes can both ask and answer, and data moves hop by hop. A blockchain full node checks and relays data itself.

How do blockchain nodes discover their peers?

They start with discovery contacts and learn additional addresses from the network. Bitcoin Core's October 2026 address table separates 1,024 new buckets from 256 tried buckets, each holding 64 entries. Bucketing by address range makes it harder for a localized attacker to fill the entire table.

How many nodes do Bitcoin and Ethereum have?

For 21 October 2026, btcnodes.io counts 25,897 reachable Bitcoin nodes, while Ethernodes counts 7,777 Ethereum execution nodes and 5,426 consensus nodes. Other crawlers produce different snapshots. ProbeLab reports 29,479 Bitcoin nodes for 20 October 2026. Reachable systems are not equivalent to users, wallets or every installed client.

What is an eclipse attack on a node?

It occurs when an attacker controls all of a target's connections and supplies a misleading network view. Bitcoin Core's optional asmap defense groups addresses by network operator. Introduced in 2020, it remains off by default in version 32.0 in October 2026. Connection counts alone cannot show whether peers are genuinely independent.

Do all blockchains broadcast transactions through global gossip?

Bitcoin and Ethereum spread transactions through gossip, reaching every reachable Bitcoin node and every Ethereum execution node. Canton does not. Only involved validators receive and confirm a transaction, while the synchronizer orders encrypted messages. Those validators store only contracts relevant to their parties.

What is a P2P crypto exchange?

It is a marketplace where people trade cryptocurrency with one another, often with escrow supporting the transaction. Custody and dispute handling depend on the platform's design. In October 2026, Bisq disputes move through trader chat, mediation and arbitration. Calling an exchange peer-to-peer does not explain which party controls escrow or handles disagreements.
Sources (35)

Get updates about blockchain, technologies and our company

We will process the personal data you provide in accordance with our Privacy policy. You can unsubscribe or change your preferences at any time by clicking the link in any email.

Follow us on social networks and don't miss the latest tech news

  • facebook
  • X
  • linkedin
  • instagram
Stay tuned and add value to your feed