We've looked at gossip protocols in the abstract. It's now time to apply those abstractions to Bitcoin's own P2P network.
At a high level, almost all cryptocurrencies inherit the same P2P network design from Bitcoin. With Gnutella as background, you should now be fully equipped to understand Bitcoin's networking layer. It's really quite similar to Gnutella, with a few augmentations that we'll cover in this lesson.
Entering the network
So far we've analyzed gossip networks at steady state. But how does one actually join a gossip network? Do you just randomly query nodes on the Internet until you find someone running the right software? (Thankfully, no.)
Every P2P protocol requires an bootstrap node to usher you into the network and help you initialize your peer list. This bootstrap node is your entrance point into the P2P network, from which point you can then organically find new peers.
In Bitcoin Core, the canonical Bitcoin implementation, these bootstrap nodes are hard-coded as trusted DNS servers maintained by the core developers.
// From: https://github.com/bitcoin/bitcoin/blob/master/src/chainparams.cpp vSeeds.emplace_back("seed.bitcoin.sipa.be"); // Pieter Wuille, only supports x1, x5, x9, and xd vSeeds.emplace_back("dnsseed.bluematt.me"); // Matt Corallo, only supports x9 vSeeds.emplace_back("dnsseed.bitcoin.dashjr.org"); // Luke Dashjr vSeeds.emplace_back("seed.bitcoinstats.com"); // Christian Decker, supports x1 - xf vSeeds.emplace_back("seed.bitcoin.jonasschnelli.ch"); // Jonas Schnelli, only supports x1, x5, x9, and xd vSeeds.emplace_back("seed.btc.petertodd.org"); // Peter Todd, only supports x1, x5, x9, and xd vSeeds.emplace_back("seed.bitcoin.sprovoost.nl"); // Sjors Provoost vSeeds.emplace_back("dnsseed.emzy.de"); // Stephan Oeste
You can retrieve an initial peer list yourself by using
dig to perform a DNS query to one of these bootstrap nodes through a UNIX command line.
$ dig seed.bitcoin.sipa.be ; <<>> DiG 9.10.6 <<>> seed.bitcoin.sipa.be ;; global options: +cmd ;; Got answer: ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 64217 ;; flags: qr rd ra; QUERY: 1, ANSWER: 25, AUTHORITY: 0, ADDITIONAL: 1 ;; ANSWER SECTION: seed.bitcoin.sipa.be. 3337 IN A 126.96.36.199 seed.bitcoin.sipa.be. 3337 IN A 188.8.131.52 seed.bitcoin.sipa.be. 3337 IN A 184.108.40.206 seed.bitcoin.sipa.be. 3337 IN A 220.127.116.11 seed.bitcoin.sipa.be. 3337 IN A 18.104.22.168 seed.bitcoin.sipa.be. 3337 IN A 22.214.171.124 seed.bitcoin.sipa.be. 3337 IN A 126.96.36.199 seed.bitcoin.sipa.be. 3337 IN A 188.8.131.52 seed.bitcoin.sipa.be. 3337 IN A 184.108.40.206 # ... and so on.
As an interesting point of history, the first version of Bitcoin bootstrapped its peer list by finding peers through IRC channels. Each Bitcoin node came bundled with a small IRC client that on first boot would join a random channel between
bitcoin99. If it found other IPs in these channels, it'd try to connect to them until it filled up its initial peer list.
This form of bootstrapping was eventually abandoned after the IRC server (LFNet) shut down, which temporarily bricked Bitcoin's peer discovery procedure. Since then, Bitcoin has relied on this DNS-based system for bootstrapping.
Initial peer discovery is an inherent chokepoint for all P2P networks. But after that phase, a node is free to populate its peer table with whichever peers it chooses.
Spam protection in Bitcoin
You'll recall that one of the weaknesses of P2P protocols is quality control. How do you stop bad actors from spamming the network to death? We hand-waved this earlier. Turns out, Bitcoin uses a reputation system to deal with this problem, implemented in 2011 by Gavin Andresen.
Say you're a Bitcoin node and you've just bootstrapped your peer list. You start by assigning each of your peers a spam score of 0. Any time that peer misbehaves toward you, their spam score goes up. Think of this like a P2P criminal record—in the absence of a central judicial system, everyone in the network has to keep tabs on everyone else's behavior.
Small infractions, such as failing to send a
version message in an initial handshake, will increase your score by 1 point. More serious DoS attempts, such as sending more than 50,000 ids in an
INV (inventory) message, will get you dinged 20 points.
Once a peer accumulates 100 points, your client automatically shadowbans them for 24 hours and stops gossiping to them. These spam scores are not actually propagated in the protocol across peers. (Why do you think they're not?) But even so, this system serves as a decent defense against misbehaving nodes.
P2P networks come with intrinsic privacy tradeoffs. Think about it: every message in the network is being shouted out in the open, and anyone is free to observe all the messages that pass by. For a financial network like Bitcoin, this is lack of privacy isn't exactly ideal.
Bitcoin affords some financial privacy thanks to its pseudonymity. If the world knows that the account
1G9HFbCRikgPpQboURsdqszy9HbKtvceZ5 has spent 0.25 BTC at a sketchy overseas pharmacy, maybe I'm okay. But if someone can figure out that my IP that originally sent the signed transaction, suddenly my privacy is compromised.
This sucks. P2P networks are resistant to censorship, but they are not at all resistant to surveillance.
Let's sketch out how a sophisticated eavesdropper might surveil the network.
By default, a normal Bitcoin nodes make 8 outbound connections to other peers. However, the Bitcoin client is open source, so anyone is free to modify their client to connect to as many peers as they want. There are about 10,000 live peers in the Bitcoin Network right now, and in principle one could connect to all of them. This would make you a supernode.
With a supernode, you could keep track of every message sent anywhere in the network and reconstruct the history of all messages from a god's-eye view. This would essentially let you figure out which addresses corresponded to which IPs and deanonymize Bitcoin transactions.
Gossip propagation is not designed to maintain privacy. Remember, the moment after a message is received, each node immediately floods it out to all of their outgoing peers. This leaves an obvious trace of where each message started. Even if a passive observer were connected to just 50% of the network, they'd be clearly see the "message wave" emanating from the sender.
In 2015, Bitcoin altered the way that it propagated gossip messages to achieve better privacy. It now uses a method called diffusion. In diffusion, instead of immediately flooding to each peer, the client waits a random exponential delay before gossiping to each of its peers. This has the effect of obscuring the P2P message graph, making it harder to observe where the "message wave" originated from.
You can sketch it in code like this:
def gossip(msg): for peer in peers: schedule_send(peer, msg, wait=np.random.exponential(1.0 / theta))
Even with this, however, privacy on Bitcoin's networking layer is far from perfect. Diffusion still leaks quite a bit of information to a passive adversary. Furthermore, P2P messages are not bilaterally encrypted, so passive packet sniffers can easily snoop on any Bitcoin traffic in cleartext.
Recently, researchers at CMU came up with an improvement over diffusion known as Dandelion Protocol that provides better network-level privacy guarantees.
In Dandelion protocol, every transaction broadcast starts with a secret game of telephone. The originator will whisper their transaction to just one peer, who whispers it to one other peer, and so on in a chain. After a random number of hops, the final peer will gossip out the transaction just as in Bitcoin. But this peer is so far removed from the originator, for any observer it’s impossible to tell who the chain started with.
This is much more effective for obfuscating the originator's IP, but it comes at the cost of slower message propagation. Dandelion is now being considered as a possible augmentation to Bitcoin's gossip mechanism, and has already been implemented in other cryptocurrencies. We've provided more resources on Dandelion in the additional reading.
Life in a P2P Network
You should now have a better mental model of how messages propagate in cryptocurrency networks. Due to their gossip architectures, cryptocurrencies are
"eventually consistent"—they don't provide any hard guarantees about when your message will be seen by the rest of the network. Furthermore, not all nodes will see the same state at the same time.
When these networks are at scale, messages can propagate quite slowly. For Bitcoin, it used to take more than 30 seconds for 90% of the network to receive the latest block! Today, it takes a few seconds until the entire network synchronizes on the latest block. (Miners see new blocks much faster, which we'll discuss later when we explore Bitcoin mining.)
This concludes our exploration of Bitcoin's networking layer. By now you should have an intuition of how gossip protocols work and what makes them so robust at scale. You also should understand why Satoshi chose to design the Bitcoin network with a P2P architecture.
With that as a foundation, in the next module we will explore consensus—first, we'll study its classical origins, leading up to Satoshi's fundamental breakthrough which made Bitcoin possible.
Take this quiz to test your learning. Once you've completed it, it's time to move on.
- Bitcoin Core 0.11: P2P Network, a Bitcoin Wiki walkthrough of Bitcoin Core's P2P networking stack (2018)
- Anonymity Properties of the Bitcoin P2P Network, showing how passive observers can use statistical techniques to deanonymize Bitcoin transactions by monitoring P2P traffic, by Giulia Fanti and Pramod Viswanath (2017)
- The Dandelion Protocol (video), a message propagation technique with stronger anonymity properties, presented by Giulia Fanti (2018)
If you've gotten this far, great work! The next chapter of the course is coming soon. And if you want me to finish it faster, please yell at me on Twitter, and my deep-seated fear of public criticism down will drive me closer to madness. ᕕ( ᐛ )ᕗ