A2.3 Data transmissions

Types of IP addressing

What an IP address does

A network can deliver data only if every device on it can be identified. Inside one LAN, devices are found by their MAC address, which is fixed to the network hardware. MAC addresses say nothing about where a device is, so they cannot be used to find a route across the world. The internet therefore adds a second, logical address that reflects which network a device belongs to.

The addresses that IP uses work like postal addresses. Part of the address identifies the network, as a postcode identifies an area, and the rest identifies one host on that network, as a house number does. A router's table needs one entry for each network, not one for every device, which keeps routing tables small enough to work on a global scale.

Two versions of IP are in use today. IPv4 dates from 1981 and is still used by most devices; IPv6 was designed in the 1990s to replace it. Both coexist on the internet, and many modern devices hold an address of each kind.

IPv4 addresses and octets

An IPv4 address is 32 bits long. Long strings of bits are hard for people to read and type, so each address is written in dotted decimal: the 32 bits are split into four groups of 8 bits, each group is converted to denary, and the four numbers are separated by dots.

Each group of 8 bits is called an octet. Because one octet has 8 bits, it can hold $2^{8} = 256$ values, from 0 (00000000) to 255 (11111111). A number above 255, such as 192.168.1.300, can never appear in a valid IPv4 address.

The diagram of the address 192.168.1.20 shows each octet converted to 8-bit binary. Every octet always takes exactly 8 bits, so small values are padded with leading zeros: 1 becomes 00000001, not 1.

With 32 bits there are

possible IPv4 addresses, about 4.3 billion. Some of them are reserved for special uses, such as private networks, testing and multicast, so fewer can be given to devices on the public internet. In the 1980s this looked like plenty; by the 2010s there were more internet-connected devices than addresses.

The conversion also works in reverse. To turn a binary address into dotted decimal, split the 32 bits into four groups of 8 and add the place values of the 1 bits in each group.

Conversion between dotted decimal and binary is a short algorithm: split the text at the dots, check that there are four parts, check that each is from 0 to 255, then convert each part to 8 bits by repeated division by 2. The Python function ipv4_to_binary follows exactly these steps, without using a built-in binary formatter.

Each pass of the loop in octet_to_binary takes the remainder after division by 2 as the next bit and places it in front of the bits found so far, so the most significant bit ends up on the left. The loop always runs 8 times, which pads small values with leading zeros. In Java, split takes a regular expression, in which a plain dot means "any character", so the dot must be written "\\."; Python's split(".") treats the dot literally. Java's / on two int values already gives integer division, where Python needs //.

Who hands out addresses

A public address must be unique across the whole internet, so addresses are handed out through a hierarchy. The Internet Assigned Numbers Authority (IANA) holds the whole address space and gives large blocks to five regional organisations. They pass smaller blocks to internet service providers and large organisations, which assign single addresses to customers and devices.

Because IPv4 has only about 4.3 billion addresses and demand kept growing, IANA's free pool ran out in February 2011, and each RIR then ran out at a different date (the case study "Running out of IPv4 addresses" gives the dates).

Three responses to exhaustion explain why the other kinds of addressing exist. Private addresses let millions of networks reuse the same address ranges internally; NAT lets many devices share one public address; and IPv6 provides so many addresses that sharing is no longer needed.

Public and private IP addresses

A public IP address is unique across the whole internet, so any router in the world can deliver a packet to it. Web servers, mail servers and the internet-facing side of every home router need one. A private IP address is chosen from a range that is set aside for use inside private networks; the same private addresses are used in millions of homes, schools and offices at once.

The private ranges are defined in a standard called RFC 1918. No routes to these ranges are carried on the public internet, and ISPs discard packets addressed to them, since the same address exists in countless separate networks and there would be no way to tell which one was meant.

Source: RFC 1918, Address Allocation for Private Internets, IETF, 1996.

The suffix after the slash is CIDR notation, covered with subnetting in subtopic A2.2: it gives the number of leading bits that are fixed for the whole range. The counts follow from the bits that remain free: /8 leaves 24 host bits, so $2^{24} = 16\,777\,216$ addresses; /12 leaves 20, so $2^{20} = 1\,048\,576$; /16 leaves 16, so $2^{16} = 65\,536$.

Private addressing has three advantages. It saves public addresses, because a whole network of devices needs only one public address between them; it lets an organisation number its devices freely without asking anyone; and devices with private addresses cannot be reached directly from the internet unless the router is set up to allow it. Its main limitation is that a private address cannot be used on the internet at all, so some form of translation, NAT, is always needed at the boundary.

Static and dynamic IP addresses

How an address is given to a device is a second, separate question. A static address is entered by hand by an administrator, or reserved permanently for one device, and does not change. A dynamic address is handed out automatically for a limited time by a DHCP server, which takes it back when the lease runs out and may give the device a different one next time.

Devices that others must find at a known address need static addresses: servers, network printers, routers and the default gateway of a LAN. If a school's file server changed its address every day, every computer configured to use it would lose its connection.

Laptops and phones that join and leave networks all day use dynamic addresses, and because leases are reused, a network can serve more visitors over a day than it has addresses. Home internet connections usually receive a dynamic public address from the ISP as well; a customer who runs a server at home may pay extra for a static one.

The two pairs of terms describe different things and combine freely. Public or private is which range an address comes from; static or dynamic is how it is assigned.

Choosing between static and dynamic assignment is a trade-off between reachability and effort. The comparison table sets out the five factors that decide it.

Network address translation (NAT)

A home network has many devices with private addresses but usually only one public address, held by the router. Network address translation is the technique that lets all of them use the internet through that one address.

The simplest form of NAT maps one private address to one public address, which saves nothing. The form used in almost every home and office router also rewrites the port number, so that many private devices can share a single public address at the same time.

The NAT diagram follows one connection from a laptop in a home network. Two devices, a laptop and a phone, both open connections from private port 51000 to the same web server; the router gives each a different public port, and the translation table records both mappings.

The process, step by step:

1. The laptop (192.168.1.10) sends a packet from its port 51000 to the web server 198.51.100.20 on port 443.

2. The router receives the packet on its LAN side. It replaces the source address with its public address 203.0.113.5, chooses an unused public port (40001) and records 192.168.1.10:51000 ↔ 203.0.113.5:40001 in its translation table.

3. The packet crosses the internet. The web server sees only 203.0.113.5:40001 as the sender.

4. The server's reply is addressed to 203.0.113.5:40001. The router looks up port 40001 in its table, rewrites the destination to 192.168.1.10:51000 and forwards the packet into the LAN.

5. When the connection has been idle for a while, the router deletes the entry so the public port can be reused.

A short program shows the logic of the translation table. The two dictionaries outbound and inbound hold the same mappings in both directions, so that the router can look up a private address and port when a packet leaves and a public port when a reply arrives.

The third call reuses the existing entry, because the laptop's connection is already in the table. The last call shows what happens to a packet that arrives for a public port with no entry: there is no private device to deliver it to, so it is dropped. In Python, next_port is changed inside a function, so it must be declared global; in Java it is a static field of the class.

Benefits and limitations of NAT

NAT has three main benefits. It saves public IPv4 addresses, because a home or a whole school needs only one; it lets private networks and the public internet communicate even though private addresses cannot be routed; and it hides the internal addresses and layout of the network from outside. As a side effect, a device on the inside cannot be contacted by an outside device that it has not contacted first, because an unsolicited packet finds no table entry. Subtopic A2.4, Network security, looks at this side effect in detail.

NAT also has limitations. It breaks the original design of the internet, in which any host could contact any other host directly: a game console or a video-calling app behind NAT cannot simply accept incoming connections, so applications need workarounds such as port forwarding (a fixed table entry set up by an administrator) or relay servers. The router must rewrite every packet and keep a table entry for every connection, which uses memory and processing time and can become a bottleneck. Many ISPs now also run carrier-grade NAT, in which thousands of customers share a small pool of public addresses, adding a second layer of translation. Using an IPv6 address for each device removes the need for any of this.

IPv6 addresses

IPv6 replaces the 32-bit address with a 128-bit one. Writing 128 bits as denary octets would need sixteen numbers, so IPv6 addresses are written in hexadecimal: eight groups of four hex digits, separated by colons. Each hex digit stands for 4 bits, so each group holds 16 bits and the eight groups hold $8 \times 16 = 128$ bits.

The number of possible IPv6 addresses is

which is about $7.9 \times 10^{28}$ times as many as IPv4. At that scale every device, sensor and appliance on Earth can have its own globally unique address, with no need for private ranges and NAT to stretch the supply.

Full IPv6 addresses are long, so two rules shorten them. Both are designed so that the full form can always be rebuilt from the short one, and the IPv6 compression diagram applies them to one address.

1. Drop leading zeros in any group: 0db8 becomes db8, 0042 becomes 42 and 0000 becomes 0. Trailing zeros must stay (ff00 stays ff00).

2. Replace one run of consecutive all-zero groups with a double colon ::. This may be done only once in an address; with two double colons there would be no way to know how many zero groups each one stands for.

To expand a shortened address, count the groups that are still written, subtract from eight to find how many zero groups the :: stands for, then pad every group back to four digits.

Comparing IPv4 and IPv6

IPv6 is a separate protocol with its own packet format. A device that understands only IPv4 cannot read an IPv6 packet. During the long changeover, most devices and networks therefore run dual stack, holding an IPv4 and an IPv6 address and using whichever the other end supports.

Comparing transmission media

Media, bandwidth and transmission time

Every link in a network carries bits as a physical signal: a changing voltage in copper wire, pulses of light in glass, or radio waves through the air. The material or space that carries the signal sets how fast data can travel, how far it can go before it must be boosted, and how easily it can be disturbed or intercepted.

The most important property of a link is how much data it can carry each second. Bandwidth is measured in bits per second (bit/s), usually with decimal prefixes: 1 Mbit/s is $10^{6}$ bits per second and 1 Gbit/s is $10^{9}$. Network speeds use a small b because they are given in bits, while file sizes are usually given in bytes (capital B), and one byte is 8 bits.

Bandwidth is not the same as latency. Bandwidth is how many bits can be sent each second, like the number of lanes on a road; latency is how long one bit takes to arrive, like the travel time from one town to the next. A satellite link can have high bandwidth but high latency, because the signal travels about 36,000 km up to a geostationary satellite and the same distance back down.

The formula gives the time to push the data onto the link. The total time for a transfer also includes the latency of the route, and in practice the throughput is lower than the bandwidth, so real transfers take longer.

Attenuation and interference

As a signal travels along any medium, it loses strength. In copper the energy is lost as heat in the wire's resistance and as radiation; in glass, light is scattered and absorbed by the material; in air, radio waves spread out and are absorbed by walls, furniture and people. When the signal becomes too weak to tell a 1 from a 0 reliably, errors appear.

Attenuation is measured in decibels (dB), usually per kilometre or per 100 m of cable. The decibel scale is logarithmic: every 10 dB of loss divides the power by 10, so a 20 dB loss leaves 1% of the signal and a 30 dB loss leaves 0.1%. Long links place repeaters or amplifiers at intervals to restore the signal before it fades below a usable level, which adds cost and points that can fail.

A signal can also be damaged by other signals. Motors, fluorescent lights, power cables and radio transmitters all produce electromagnetic fields, and a copper wire picks these up as unwanted voltages, just as an aerial picks up radio.

Inside a cable that holds several pairs of wires, the signal in one pair can leak into its neighbours. This form of interference grows with the frequency of the signals and with the length of cable that the wires run side by side.

Both attenuation and interference reduce the usable bandwidth and range of a medium, because the receiver can only decode a signal that stands clearly above the noise. They are the main reasons why the three media differ.

Twisted-pair cable

The cable that connects most desktop computers, printers and access points to a switch is twisted-pair copper cable, often just called an "Ethernet cable" because it carries Ethernet. It contains four pairs of thin insulated copper wires, and the two wires in each pair are twisted around each other along the whole length of the cable.

Twisting the wires lets the cable carry fast signals in spite of interference. The two wires of a pair carry the same signal with opposite polarity, and the receiver measures the difference between them. Interference from outside reaches both wires almost equally, because twisting keeps swapping which wire is closer to the source, so it shifts both voltages by the same amount and disappears when the difference is taken. Each pair is twisted at a different rate, so neighbouring pairs never run in parallel for long, which reduces crosstalk.

The cross-sections of the two kinds of cable show their layers. A twisted-pair cable holds four pairs of copper conductors, each wire in its own plastic insulation; a fibre-optic cable has a glass core surrounded by cladding, a protective coating and an outer jacket.

Twisted-pair cable is made in categories. Higher categories are built more precisely, with tighter twists and sometimes a foil shield, so they carry higher frequencies and therefore more data.

Source: Category 5 and Category 6 cable standards (TIA), summarised on Wikipedia, accessed October 2026.

The 100 m limit is set by attenuation and the timing rules of Ethernet: beyond it, the signal is too weak and distorted to decode reliably, so a switch is needed every 100 m. Twisted pair is cheap, light and flexible, and it is easy to install: a technician can cut it to length and fit the plug (an RJ45 connector) on site with a hand tool. Most cable is unshielded (UTP); shielded versions (STP) wrap the pairs in foil for factory floors with heavy motors, at extra cost.

Its weaknesses all come from carrying electricity. Copper picks up EMI near motors, lift machinery or power lines, it suffers from crosstalk, and its 100 m range makes it unsuitable for links between buildings. A cable also gives off a weak electromagnetic field and can be tapped by anyone with physical access to a socket or switch, so wired networks still need physical security.

Fibre-optic cable

Each strand of very pure glass is thinner than a human hair. A laser or LED at one end switches on and off billions of times a second, and a photodetector at the other end turns the light pulses back into bits.

The light stays inside the glass because of total internal reflection. The core is surrounded by cladding made of glass with a slightly lower refractive index. When light in the core hits the boundary at a shallow angle, it is reflected back in completely, so it zig-zags (or, in thin single-mode fibre, travels almost straight) along the core even when the cable bends gently, as the reflection diagram shows. Two types are common:

• Multimode fibre has a wider core that lets light take several paths. It is cheaper to connect but the paths arrive at slightly different times, which smears the pulses, so it is used for links up to a few hundred metres inside buildings and data centres.

• Single-mode fibre has a core only about 9 micrometres across, so light takes one path. It carries signals for tens of kilometres between amplifiers and is used for long-distance, undersea and ISP links.

Fibre's strengths come from using light in glass. Its attenuation is extremely low, so signals travel far without boosting, and it can carry very high frequencies, giving huge bandwidth; a single fibre can carry many wavelengths of light at once, each a separate channel. Light is not affected by electromagnetic fields, so fibre is immune to EMI and crosstalk, and can run beside power cables or through factories. It gives off no electromagnetic signal, so it cannot be read from outside; tapping it means physically bending or cutting the glass, which causes a loss the network equipment can detect.

Fibre's limitations are cost and handling. The cable itself is now cheap, but the lasers, receivers and switches with optical ports cost more than copper equipment. Joining two fibres (splicing) needs the glass ends to be cleaved perfectly flat and aligned to a fraction of a micrometre, which requires trained technicians and special tools. Glass can crack if the cable is bent too tightly or crushed, and the electronics at each end still need electrical power, which fibre cannot carry.

Low attenuation is the reason fibre works over long distances. The total loss on a link is the loss per kilometre multiplied by the length, and the decibel rule gives the fraction of power left.

The fall in attenuation was the breakthrough that made fibre networks possible. In 1965 Charles Kao and George Hockham argued that impurities, not the glass itself, caused most of the loss, and that fibre would be practical below 20 dB/km; Corning reached 17 dB/km in 1970, and modern low-loss single-mode fibre loses 0.148 dB/km at a wavelength of 1550 nm, so that 71% of the light remains after 10 km. Kao received the Nobel Prize in Physics in 2009 for this work.

Source: "Optical fiber", Wikipedia, citing Corning, accessed October 2026.

Wireless transmission

Wireless transmission carries data as electromagnetic waves through the air, with no cable at all. In a Wi-Fi network, devices exchange radio signals with a wireless access point connected to the wired LAN; a mobile network covers a wide area with cells, each served by a mast; and satellites relay signals to places with no cables.

Wireless has two big advantages: mobility, because devices can move freely while connected, and easy installation, because adding a device needs no cable through walls or ceilings. An access point can serve a classroom, a café or a lecture hall full of phones that would be impossible to cable, and a temporary event can be networked in an hour.

Its disadvantages come from sharing open space. All the devices on one access point share the same radio channel and largely take turns to transmit on it, so the bandwidth is divided among users and each device's share falls as more join. Radio signals weaken quickly with distance and are absorbed by walls, floors, water and metal, so the range is a few tens of metres indoors. Wi-Fi in the 2.4 GHz band shares its frequencies with Bluetooth, microwave ovens and neighbouring networks, so interference is common. The newest standard, Wi-Fi 7, has a theoretical maximum of 23 Gbit/s in a single band, but real throughput is much lower because of distance, walls, other users and interference.

Wireless is also the least secure medium. Radio signals spread in every direction, well beyond the walls of a building, so anyone within range can receive them without touching the network. Without strong encryption, data can be read by an eavesdropper; without strong authentication, outsiders can join the network. Subtopic A2.4, Network security, covers the protections used on Wi-Fi.

Comparing the three media

The table compares the three media across the eight factors used to choose between them. Values are typical for current equipment; any of them can change with the exact standard used.

No medium is best on every factor, which is why real networks combine them. A typical school uses single-mode or multimode fibre between buildings and between the server room and each floor (the backbone), twisted pair from each floor's switch to the desks and access points, and Wi-Fi for laptops and phones. The worked example applies the same reasoning to a choice between two media.

Packet switching

Why data is split into packets

A file, an email or a video stream is far too large to send as one block. If a 1 GB file were sent in one piece, it would hold a link for the whole transfer, every other user would have to wait, and a single corrupted bit near the end would mean sending the whole gigabyte again. Data sent over the internet is therefore split into small pieces, each sent separately.

Each packet is a self-contained unit with everything a router needs to forward it. On the internet a packet is usually at most about 1,500 bytes, the largest unit an Ethernet link carries in one go, so a 5 MB photo becomes several thousand packets.

The data itself travels behind the header. Only the hosts at each end look inside it; the routers in between read only the header.

The packet structure diagram shows the three parts of the packet model used to explain packet switching: a header, a payload and a trailer holding an error check.

The trailer holds a checksum, a short value calculated from the bits of the data. The receiver repeats the calculation on what arrived; if the two values differ, some bits were changed on the way, and the damaged data is discarded and sent again.

On the real internet these fields are spread across the layers of the TCP/IP model. The IP header holds the addresses and the time to live; the sequence number belongs to the TCP header inside the payload of the IP packet; and the error-checking trailer is added to each frame on each link, with IPv4 also carrying a checksum of its own header. The three-part model brings these fields together in one simple picture of a packet.

How packet switching works

Packet switching is the method the internet uses to move data. It works like sending a long letter as many separate postcards, each numbered and each addressed to the same house: every postcard goes through the postal system on its own, and the receiver sorts them by number.

The process for one transfer, step by step:

1. Splitting. The sending host splits the data into pieces small enough to fit in a packet.

2. Adding headers. Each piece is given a header with the source and destination IP addresses, a sequence number showing its position in the data, a time to live, and other control information. A checksum is calculated so that damage can be detected.

3. Sending independently. Each packet is passed to the first router (the default gateway). From then on it travels on its own, and the network keeps no record that it belongs with the others.

4. Routing at each hop. Every router that receives a packet reads its destination address, looks it up in its routing table and forwards it on the link to the next router. Different packets of the same message may be sent different ways, for example when a link becomes congested or fails.

5. Arriving. Packets can arrive in a different order from the one in which they were sent, and some may be lost or damaged.

6. Reassembling. The receiving host holds the packets in a buffer, uses the checksums to discard damaged ones, puts the rest in order by sequence number, requests any missing ones again, and joins the payloads back into the original data.

The routes diagram follows three packets of one message across a small network. Packet 3 takes a longer path through R5 and packet 1 is delayed in a queue at R2, so packet 2 arrives first. Because each packet carries its sequence number, the receiver can still rebuild the message correctly. Moving each packet towards its destination, one router at a time, is called routing.

Each router holds a table of the networks it knows how to reach. For every destination network, the table gives the next router to send packets to (the next hop) and the interface to send them out of. A router's table does not hold the whole route, only the best next step; the next router's table holds the step after that.

A simplified routing table for a school's router shows the idea. The last row, the default route 0.0.0.0/0, matches every address that no other row matches, and sends it to the ISP.

Each step from one router to the next is called a hop. A packet travelling between continents, from a laptop in London to a server in Singapore for example, often crosses a dozen or more hops.

A mistake in routing tables can send packets round in a loop, from router A to B to C and back to A, for ever. To stop such packets clogging the network, every packet carries a counter. Each router that forwards the packet subtracts 1, and a router that reduces it to 0 discards the packet and sends an error message back to the sender. Common starting values are 64 and 128.

At the destination, the receiving host rebuilds the data from the packets that have arrived.

In the internet, reassembly and recovery of lost packets are carried out by TCP on the receiving host. TCP acknowledges what has arrived; if the sender gets no acknowledgement for some data in time, it sends that data again. The routers in between do none of this: they forward each packet as best they can and drop it if their queues are full. A short Khan Academy lesson with Vint Cerf, one of the designers of TCP/IP, follows a message as it is broken into packets, routed independently and checked for completeness on arrival.

The program make_packets splits a short message into packets with 6-character payloads, each stored as a sequence number and a payload. The function reassemble then puts packets that arrived in any order back into slots by sequence number, and reports any that are missing.

The packets arrive in the order 2, 0, 3, 1, yet the message is rebuilt correctly because each payload is placed in the slot given by its sequence number, not in the order it arrived. In the second run packet 2 never arrives, so its slot stays empty and the receiver requests it again. Java has no simple built-in pair type, so the Java version stores each packet as a two-element String array and works out the number of packets with integer division rounded up, $(24 + 6 - 1) / 6 = 4$.

Splitting data into packets has a cost: every packet carries its own header, so more data is sent than the file itself, and the overhead grows as the packet size falls.

The roles of switches and routers

Two kinds of device move packets on their way, and each works at a different scale. A switch moves data within one LAN; a router moves packets between networks.

Inside a LAN, the switch reads the destination MAC address of each frame and sends it only out of the port leading to that device, using a table of which MAC address is reached through which port. It does not read IP addresses and holds no information about networks beyond the LAN. When a packet must leave the LAN, the host sends it to the MAC address of its default gateway, the router.

A router reads the destination IP address in the packet header, finds the matching network in its routing table and forwards the packet towards the next hop. Before sending it on, the router subtracts 1 from the TTL and wraps the packet in a new frame with new source and destination MAC addresses for the next link. The IP addresses stay the same from one end of the journey to the other (unless NAT rewrites them at the edge), while the MAC addresses change at every hop.

Circuit switching and packet switching

The telephone network that came before the internet worked differently. Before a call could start, the exchanges reserved a complete path from caller to receiver, and that path was held for the call's whole length, whether anyone was speaking or not.

The diagram of the two methods uses the same small network, whose switching nodes S1 to S4 stand for telephone exchanges in a circuit-switched network and for routers in a packet-switched one. With circuit switching, capacity on every link of the path from A to B is held for the call, even in silences; with packet switching, packets from A and from C take turns on whatever links are free.

Packet switching suits computer data because it is bursty: a browser sends nothing for seconds, then a page's worth of requests at once. Holding a circuit for each user would leave most of the capacity idle, whereas packets from thousands of users can fill the same link continuously. It is also resilient, because no path is fixed in advance: when a router or link fails, the remaining routers send packets another way. Its weakest point is variable delay (jitter), which matters for live voice and video; video-calling applications smooth it with small buffers, and networks can give such traffic priority in their queues.

Static and dynamic routing (HL only)

Where routing tables come from

A router knows its directly connected networks as soon as its interfaces are given addresses: a router with one interface on 192.168.1.0/24 and another on a link to a second router can reach both without being told. Every other network in its routing table must be added in one of two ways. An administrator can type the routes in by hand, or the routers can work them out for themselves by exchanging information. These two methods are static and dynamic routing.

The triangle network is used for both methods. Three routers each serve one LAN, and the routers are joined in a ring, so there are always two possible paths between any two LANs.

Static routing

In static routing, a network administrator enters each route into each router by hand. For every network that is not directly connected, the administrator types the destination network, its mask and the next hop. The router uses exactly those routes until someone changes them.

For the triangle network, the administrator might give R1 these routes. The first row appears automatically; the other two are typed in.

The process of static routing, step by step:

1. The administrator plans the network and decides the best path from every router to every network.

2. On each router, the administrator enters one static route for every network that is not directly connected, giving the destination network and the next hop.

3. When a packet arrives, the router matches its destination address against the table and forwards it to the next hop given.

4. If the network changes (a new LAN is added, a link fails or is moved), nothing happens automatically: the administrator must log in to every affected router and edit its routes.

Static routing has clear advantages. It uses no processor time, memory or bandwidth for exchanging routes, so even a small, cheap router can handle it. The routes are completely predictable: the administrator knows exactly which way every packet goes. It is also more secure, because the routers accept no routing information from other devices, so an attacker cannot inject false routes to redirect traffic.

Its disadvantages grow with the size of the network. Every route must be typed in, and in a network where each of $n$ routers serves one LAN and must reach the other $n-1$, the administrator has to enter $n(n-1)$ routes. Typing errors can create loops or black holes, where packets are sent to a router with no onward route. Most importantly, static routes do not adapt to failures: if link 3 fails, R1 keeps sending LAN C's packets towards R3 over the dead link, and they are lost until the administrator notices and changes the route, even though a working path through R2 exists all along.

A common compromise is the floating static route: a second, back-up static route to the same network through a different next hop, marked as less preferred. The router uses it only if the main route's link goes down. It adds some resilience but still covers only the failures the administrator thought of in advance.

Dynamic routing

In dynamic routing, the routers build and update their own routing tables. Each router runs the same routing protocol, finds its neighbours, and regularly sends them a list of the networks it can reach and the cost of reaching each. From the information it receives, each router works out the best path to every network.

The rules for this exchange are set by a routing protocol. The method can be understood without the details of any one protocol, but their names appear in real networks: RIP and OSPF are used inside an organisation's network, and BGP is used between the networks that make up the internet.

The process of dynamic routing, step by step:

1. Configuration. The administrator turns on the routing protocol on each router and tells it which of its networks to advertise. No individual routes are typed in.

2. Discovery. Each router finds its neighbouring routers by sending small "hello" messages on its links.

3. Exchange. Neighbours send each other the networks they can reach and a metric (cost) for each, such as the number of hops or a value based on the bandwidth of the links.

4. Calculation. Each router selects, for every network, the path with the lowest total metric and puts its next hop in the routing table.

5. Adapting. Routers keep exchanging updates. When a link fails, the routers at each end detect it (the interface goes down, or hello messages stop arriving) and send updates; every other router recalculates its routes, and traffic moves to the best remaining path.

In the triangle network, if link 3 fails, R1 and R3 detect it within seconds and send updates to R2, and all three recalculate. R1's route to LAN C changes from "R3 over link 3" to "R2 over link 1", and R2 forwards those packets over link 2. No one has to touch the routers. The time this takes is the convergence time.

The convergence timeline follows the triangle network after link 3 fails. Until all the routers have converged, their tables disagree, so some packets are sent towards the failed link or round a loop and are lost.

Dynamic routing's advantages are that it adapts automatically to failures and changes, finding a new path without human action, and that it scales: adding a new LAN means advertising it on one router, and every other router adds it to its table automatically. On a network with several possible paths, it can also choose the best one by its metric and move traffic away from slow or broken links.

Its disadvantages are cost and complexity. Every router spends processor time and memory running the protocol and storing the routes it receives, and the update messages use bandwidth on every link. The protocols are complex to set up and troubleshoot, and a mistake can spread: a router that advertises a wrong route can mislead all the others. Updates from outside must also be protected, because a device that sends false routing information can redirect traffic. During convergence, routes are inconsistent and packets can be lost.

Comparing static and dynamic routing

The table compares the two methods across the seven factors used to choose between them.

Static and dynamic routing are often combined. A home or small office router uses a single static default route to its ISP; there is only one way out, so dynamic routing would add nothing. The ISP and the internet's core use dynamic routing, because their networks are huge, constantly changing and full of alternative paths.

Case study: Running out of IPv4 addresses

Early allocations of IPv4 addresses were generous. Organisations that joined in the 1980s and early 1990s, mostly in North America and Europe, often received very large blocks of addresses, some with millions of addresses for a single company or university. As home broadband, mobile phones and connected devices spread around the world, demand rose far beyond the 4.3 billion addresses that 32 bits allow, and the shortage was spread unevenly, hitting hardest the regions that came online later and grew fastest.

The central pool held by IANA ran out first. On 3 February 2011, IANA gave its last five blocks of 16.8 million addresses (each a /8), one to each regional internet registry. Each RIR then ran out on its own date, depending on how fast demand was growing in its region.

APNIC, the Asia–Pacific registry, ran out first, about ten weeks after IANA, because its region had the fastest-growing number of internet users and had received fewer addresses in the early years. Since then, a new network in most of the world can obtain public IPv4 addresses only by buying or renting them from an organisation that has spare ones, or by waiting for recovered addresses.

Source: "IPv4 address exhaustion", Wikipedia, accessed October 2026. The RIRs still hand out small numbers of recovered addresses.

How the internet kept working

Three stopgaps and one replacement kept the internet growing despite the shortage, as the table of responses sets out. Private addresses and NAT with PAT carried most of the load. NAT was first described in 1994 as a "short-term solution" to address depletion, and more than thirty years later it is still in almost every home router, with ISPs adding carrier-grade NAT on top.

Source: RFC 1631, The IP Network Address Translator (NAT), IETF, May 1994.

The long-term answer is IPv6, and its adoption has been slow but steady. IPv6 traffic to Google passed 50% of the total for the first time on 28 March 2026, and in April 2026 the share was about 45–50% depending on the day of the week. Adoption differs greatly between countries and between ISPs, so some countries send most of their traffic over IPv6 while others send almost none.

Source: Google IPv6 statistics, reported in "IPv6 deployment", Wikipedia, accessed October 2026.

Evaluating the response

The case shows the main trade-off of NAT. It solved the immediate shortage cheaply, with no change to the devices behind the router, at the price of the internet's original property that any device could be reached directly, so peer-to-peer applications, online games and home servers have needed workarounds ever since.

It also shows why replacing a protocol is so hard. IPv6 is not directly compatible with IPv4, so every device, router, application and ISP has to support both during the change, which costs money and offers little benefit to any one organisation while NAT still works. The changeover therefore depends on many independent organisations acting together, and after about three decades it is only half complete.

Exam focus

Data transmissions is examined in Paper 1, Section A, in structured questions built around a scenario such as a school, hotel or small business network. Short parts ask you to state a fact (the length of an IPv6 address, a private range), identify a private address or calculate a transmission time or a binary conversion, for 1–2 marks each. Describe parts on IP addressing and NAT need the features or steps in order, for 2–4 marks. Compare parts on media need both media referred to in every point, usually for 4–6 marks, and explain parts on packet switching need the stages with the reason for each, for 4–6 marks. HL papers add parts on static and dynamic routing, usually "explain an advantage and a disadvantage" or an evaluation for a given network, judged against the factors of configuration, maintenance, complexity, resource usage, convergence, scalability and network size.

Vague answers such as "NAT hides your IP" or "the router sends it" lose most marks on IP addressing, media and packet switching. Every answer should name the mechanism (32 bits against 128 bits, low attenuation, a translation table, a routing table look-up) and, for explain and compare questions, give its consequence.

Practice questions

1. State two private IPv4 address ranges. [2]

2. Identify whether each of these addresses is public or private: 172.32.4.1, 10.200.0.1, 192.168.255.1. [3]

3. Convert the binary address 00001010.00000001.11001000.00000011 to dotted decimal. [2]

4. Write the IPv6 address fe80:0000:0000:0000:0000:0000:1a2b:3c4d in its shortest form. [2]

5. Distinguish between a static IP address and a dynamic IP address, and state one device that should have a static address. [3]

6. Explain why NAT is needed by a home network that uses IPv4. [3]

7. Calculate the minimum time to transmit a 25 MB file over a link with a bandwidth of 20 Mbit/s (1 MB = $10^{6}$ bytes). [2]

8. Outline the purpose of the time to live (TTL) field in a packet header. [2]

9. Compare circuit switching and packet switching. [4]

10. (HL only) Explain the meaning of convergence in dynamic routing, and why a long convergence time is a problem. [3]

Summary

• An IP address is a logical address that identifies a device and its network. IPv4 uses 32 bits written as four denary octets (about 4.3 billion addresses); IPv6 uses 128 bits written as eight groups of four hex digits (about $3.4 \times 10^{38}$ addresses), shortened by dropping leading zeros and replacing one run of zero groups with ::.

• Public addresses are globally unique and routable on the internet; private addresses (10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16) are reused inside private networks and never routed on the internet.

• Static addresses are configured manually and stay the same (servers, printers, routers); dynamic addresses are leased automatically by DHCP and can change (laptops, phones).

• NAT, usually as PAT, rewrites private source addresses and ports to one public address, records each mapping in a translation table and maps replies back, saving IPv4 addresses. It is not a firewall.

• Bandwidth is the maximum data rate of a link in bits per second; transmission time = data size in bits ÷ bandwidth. Attenuation is signal loss over distance; EMI and crosstalk are unwanted signals.

• Fibre-optic cable has the highest bandwidth, lowest attenuation and longest range, is immune to EMI and hard to tap, but costs most and is hardest to install. Twisted pair is cheap and easy but limited to about 100 m and affected by EMI. Wireless is mobile and easy to deploy but has short range, shared bandwidth, high interference and the weakest security.

• In packet switching, data is split into packets, each with a header (addresses, sequence number, TTL) and payload; packets are routed independently, hop by hop, using routing tables, may arrive out of order and are reassembled by sequence number at the destination.

• Switches forward frames within a LAN using MAC addresses; routers forward packets between networks using IP addresses and routing tables. Circuit switching reserves a dedicated path; packet switching shares links and survives failures.

• (HL) Static routes are typed in by an administrator: simple, secure and resource-free, but they do not adapt to failures and scale poorly. Dynamic routing protocols share reachability between routers and recalculate routes automatically, scaling to large networks and converging after failures, at the cost of resources and complexity.

Self-check

• To do:

  I can convert an IPv4 address between dotted decimal and 32-bit binary.

• To do:

  I can work out how many addresses IPv4 and IPv6 provide from their number of bits.

• To do:

  I can shorten and expand an IPv6 address using the two compression rules, and explain why :: may appear only once.

• To do:

  I can identify whether an IPv4 address is private, including addresses at the edges of the 172.16.0.0/12 range.

• To do:

  I can distinguish public from private addresses and static from dynamic addresses, with a suitable device for each.

• To do:

  I can describe step by step how NAT with PAT translates an outgoing packet and its reply, using a translation table.

• To do:

  I can explain why NAT saves public IPv4 addresses and why it is not a firewall.

• To do:

  I can calculate a transmission time, converting bytes to bits and using powers of ten for prefixes.

• To do:

  I can explain how twisting reduces interference in twisted-pair cable and how total internal reflection keeps light in a fibre.

• To do:

  I can compare fibre-optic cable, twisted-pair cable and wireless across all eight factors, applied to a scenario.

• To do:

  I can describe the header, payload and trailer of a packet and the purpose of each field.

• To do:

  I can explain each stage of packet switching, including why packets arrive out of order and how they are reassembled.

• To do:

  I can trace the TTL of a packet across a series of routers.

• To do:

  I can distinguish the roles of switches and routers, and compare circuit switching with packet switching.

• To do:

  I can construct a static routing table for a small network and evaluate static and dynamic routing for a given network, including convergence (HL only).