How packets find their way

Each router lowers a packet's TTL by one. When it reaches 0 the packet is thrown away, so a packet caught in a loop cannot circle forever.

Try the simulator before reading anything else. Set TTL to 2 and press Send, then set it to 64 and send again. Watch where the packets stop the first time.

What you just saw: a countdown in every packet

With TTL 2, every packet died at the second router it met and a red message came back. With 64, the packets reached the server with 60 left after four routers. The IPv4 header has an 8-bit field called Time to Live (TTL), so its largest value is 255. RFC 791 says every router must lower it by at least one, and many computers start it at 64. When it reaches 0 the router drops the packet and sends an ICMP "Time Exceeded" message (RFC 792) back to the sender. That was the red message.

Why have a countdown at all? A mistake in the routing tables could send a packet round in a circle, and without a limit it would use links forever. The name says "time", and on paper RFC 791 measures it in seconds. But every router takes off at least one, so in practice it works as a count of hops. IPv6 gave the same field a plainer name, Hop Limit.

The traceroute button uses this rule to draw a map. It sends a probe with TTL 1: the first router drops it and replies, which reveals that router's address. TTL 2 reveals the second router, and so on until the destination answers. Press it and read the list it builds. Then cut R1 to R3 and send again to see the packets take another way.

Each router looks up one line in a table

No router knows the whole route from your laptop to a server on another continent. Each one makes a single decision: for this destination address, which neighbour gets the packet next? The packet moves one hop, and the next router decides again. A route is a chain of these local choices.

The decision comes from a forwarding table, a list of prefixes, each with the neighbour to use. The router picks the entry that matches the destination with the longest prefix, because a longer prefix is a more exact match. Say the table sends 203.0.0.0/16 to neighbour B and 203.0.113.0/24 to neighbour A. A packet for 203.0.113.9 matches both, the /24 wins, and it goes to A. A packet for 203.0.7.1 matches only the /16, so it goes to B. Most tables also have a default entry for everything else; in a home router it points to the ISP.

Who writes the tables

Inside one company, routers share what they can reach using routing protocols and work out new paths when a link fails. Between the tens of thousands of separate networks that make up the internet, the Border Gateway Protocol (BGP, RFC 4271) carries announcements such as "I can reach 198.51.100.0/24". When a cable is cut, routes through it are withdrawn and traffic shifts to whatever path is left. In the simulator, that is what happened when you cut R1 to R3.

Most lost packets fall out of a full queue

Routers also make packets wait. At busy moments packets arrive faster than an outgoing link can carry them, so each router keeps a queue in memory. When the queue is full, the next packet to arrive is simply dropped. Nobody is told; the sender can only work it out later. This is the most common way packets go missing on the internet, more common than a cut cable. A weak Wi-Fi signal can also spoil a frame so that it has to be thrown away. IP itself promises nothing. RFC 791 says it does not provide a reliable communication facility, with no acknowledgements and no retransmissions. Getting lost data back is left to the computers at the two ends, and that is the next unit.

A 3,000,000-byte photo is 2055 packets

Links limit how big one packet may be. On Ethernet one frame carries at most 1500 bytes of payload (RFC 894). Take away 20 bytes of IPv4 header and 20 bytes of TCP header, and about 1460 bytes of your data fit in each packet. A 3,000,000-byte photo therefore needs 3,000,000 ÷ 1460 ≈ 2054.8, rounded up to 2055 packets, and each one is routed on its own.