What travels on the wire
A message crosses the internet as numbered packets, each wearing an envelope that every box on the way can read; this page shows what is written on that envelope and who writes it down.
When you open a web page, nothing travels to you as one piece. Your request and the reply are cut into small parcels, and each parcel carries a label saying where it came from and where it is going. Most of what the rest of this site calls “hiding” is about that label: who can read it, and who keeps a copy. This page explains the label and the machines that read it, from the laptop on your desk to the server on the other side.
The addresses in the figures come from blocks reserved for documentation, so none of them belongs to a real network.1 They follow one scene throughout: a person whose home connection has the public address 198.51.100.23 logs in to a forum at forum.example (server 203.0.113.10) at 03:14:07 UTC on 14 March 2026, as the account anon_4471.
A message becomes packets
Networks carry data in packets of limited size. On Ethernet, the most common kind of local network, an IP packet can be at most 1,500 bytes long.2 A web page, a photo or even a long form submission is larger than that, so the sending computer cuts it into slices, wraps each slice in a header and sends the slices one after another. The receiver puts them back in order.
Two consequences follow. First, every packet carries the addresses, not just the first one, so anyone who sees any part of the conversation sees who is talking to whom. Second, the packets are independent: each one is forwarded on its own, and a network does not need to understand the message to deliver it.
The envelope: the IP header
The header the Internet Protocol puts in front of each slice is defined in RFC 791, published in 1981 and still the format of every IPv4 packet.3 It is 20 bytes long in the usual case and holds a dozen fields. Two matter most here: the destination address, which every router reads to decide where to send the packet next, and the source address, which tells the other side where to send replies.
The comparison with a letter is close. The header is the outside of the envelope: postal workers have to read it, and they can copy it down. The data is the letter inside. With HTTPS the letter is sealed by TLS, but the envelope cannot be sealed, because nobody could deliver it. The time-to-live field (TTL) is a hop counter: each router lowers it by one and discards the packet when it reaches zero, so a packet caught in a routing loop does not circle forever.3
IPv6, the newer version, uses a different header with longer addresses, but the same two fields sit in it, in the clear, for the same reason.4
The same packet can also be read as a record with five fields, the ones the figure on the home page uses: who sent it, where it goes, what it says, when and how much, and which account it belongs to. Encryption changes only some of them.
Hop by hop
No single machine knows the whole route. Each router looks at the destination, picks the neighbour that is closer to it, and hands the packet on. Between your laptop and a server there are several such hands: your home router, several routers inside your ISP, often an internet exchange where networks meet, and the routers of the server’s own network.
Your laptop builds the packet. Its source address 192.168.1.10 is private: it means something only inside your home, and thousands of other homes use the same number.
Forwarding a packet does not require keeping a record of it, and most routers keep none per packet. Records come from somewhere else. The ISP keeps track of which customer line was given which public address, because it needs that to run the service and bill for it. Networks can also export flow records, one line per conversation with the addresses, ports, start and end time and byte counts, using formats such as IPFIX.5 Exchanges and large networks can also sample a small fraction of packet headers with sFlow to see where their traffic goes.6 And the server at the end writes an access log line for each request, with the address and the time. That last line is where most investigations start, as the records page shows.
Private addresses and NAT
There are not enough IPv4 addresses for every device. Three blocks, 10.0.0.0/8, 172.16.0.0/12 and 192.168.0.0/16, are set aside for private use and are never routed on the public internet; any home or office may use them internally.7 The laptop’s 192.168.1.10 is such an address. Thousands of other homes use the same number.
To reach the internet, the home router performs network address translation (NAT). It replaces the private source address with its own public address and a port number it chooses, and keeps a table so that replies get back to the right device. Every device in the house shares one public address.
Many ISPs now do the same thing one level up. With carrier-grade NAT (CGN), customer routers get addresses from a block reserved for this purpose, 100.64.0.0/10, and the ISP translates them to a smaller pool of public addresses shared by many customers at once.8 In October 2017 Europol said that CGN was used for 90% of mobile internet access and 50% of fixed-line access, that one address could be shared by several thousand people, and that mobile providers using it often could not tell which subscriber had used an address.9
The fix is in the logs, not the network. The IETF’s logging recommendation for servers says that, because addresses are shared, a server should log the source port and an accurate timestamp along with the address.10 The ISP, for its part, has to log which port range each customer was given when. If either side is missing, the trail stops at a crowd.
Names: the DNS lookup
People type names, packets need numbers. The Domain Name System (DNS) translates one to the other.11 Your device asks a resolver, usually run by your ISP or a public service, and the resolver asks the hierarchy on your behalf: the root servers, then the servers for the top-level domain, then the server for the domain itself. With query minimisation, the resolver tells each server only as much of the name as that server needs to answer.12
1. Your laptop asks the resolver for forum.example. The question comes from your home address, 198.51.100.23, so the resolver learns who asked for which name. Unless the query is encrypted, your ISP can read it on the way.
The privacy point is who learns the full name together with your address: the resolver, always, and anyone on the path if the query travels in plain text, which classic DNS does. DNS over TLS and DNS over HTTPS encrypt the query between you and the resolver.136 That hides the name from the ISP’s wire, but not from the resolver, which now becomes the party that holds the record.
Ports and connections
An address gets a packet to a machine; a port number gets it to the right program on that machine. Servers listen on well-known ports, such as 443 for HTTPS, and the client picks a temporary port for its side of each connection; the IANA registry records which numbers are assigned to which services.14 Four numbers together, source address, source port, destination address and destination port, name one conversation. That is why the source port matters so much behind NAT: it is the only number that tells two customers on the same public address apart.
TCP, the protocol most web traffic uses, sets up each connection with three packets before any data flows: the client sends SYN, the server answers SYN-ACK, the client confirms with ACK.15 It numbers the bytes it sends so the receiver can put slices back in order and ask for missing ones again. None of this is encrypted. TLS protects the data inside the connection, not the fact of the connection.
What an ISP sees without breaking encryption
Put the pieces together and look from the ISP’s position. It does not need to break anything. It sees which of its customers sent each packet, the server’s address, and usually the name: once in the DNS query, unless DNS is encrypted, and once more in the opening message of TLS, where the client states which site it wants in a field called Server Name Indication (SNI), sent in clear text.16 A newer extension, Encrypted Client Hello, hides that name from the path when both browser and site support it.7 The ISP also sees when each connection starts and ends and how many bytes go each way.
The IETF’s own analysis of pervasive surveillance puts it plainly: encrypted traffic still exposes everything in its network and transport headers, addresses and ports included, and an observer who cannot decrypt anything can still fingerprint encrypted sessions and draw inferences from them.17 Sizes and timing are what website-fingerprinting research uses to guess which page is loading, even through Tor.24
How this maps to the chapters
The network chapter is about one field in the header: the source address. VPNs, proxies and Tor all work by making that field belong to someone else, which means handing your packets to a relay that sees what your ISP used to see. The question becomes who that relay is and what it writes down.
The device chapter is about what happens inside the encrypted letter and on the machine that writes it. A perfect envelope does not help if the program on the device reveals the real address another way, or if the device itself is identifiable.
The records page picks up the trail from the other end: the log line at the server, and the chain of records that turns an address and a time into a name.
Sources
- IETFRFC 5737: IPv4 Address Blocks Reserved for Documentation
- IETFRFC 894: A Standard for the Transmission of IP Datagrams over Ethernet Networks
- IETFRFC 791: Internet Protocol
- IETFRFC 8200: Internet Protocol, Version 6 (IPv6) Specification
- IETFRFC 7011: Specification of the IP Flow Information Export (IPFIX) Protocol for the Exchange of Flow Information
- IETFRFC 3176: InMon Corporation's sFlow: A Method for Monitoring Traffic in Switched and Routed Networks
- IETFRFC 1918: Address Allocation for Private Internets
- IETFRFC 6598: IANA-Reserved IPv4 Prefix for Shared Address Space
- EuropolAre you sharing the same IP address as a criminal? Law enforcement call for the end of Carrier Grade NAT (CGN) to increase accountability online
- IETFRFC 6302: Logging Recommendations for Internet-Facing Servers
- IETFRFC 1034: Domain Names - Concepts and Facilities
- IETFRFC 9156: DNS Query Name Minimisation to Improve Privacy
- IETFRFC 7858: Specification for DNS over Transport Layer Security (TLS)
- IETFRFC 6335: Internet Assigned Numbers Authority (IANA) Procedures for the Management of the Service Name and Transport Protocol Port Number Registry
- IETFRFC 9293: Transmission Control Protocol (TCP)
- IETFRFC 6066: Transport Layer Security (TLS) Extensions: Extension Definitions
- IETFRFC 7624: Confidentiality in the Face of Pervasive Surveillance: A Threat Model and Problem Statement