Standard definitions from introductory networking courses built on the top-down approach (Kurose and Ross, chapter 1). Every number in the worked examples was recomputed in code before publishing, and the probability figure was computed exactly from the binomial distribution.
Read the picture first, then the vocabulary. Work both calculations on paper before you look at the steps. They are short, but they are exactly the kind of question that turns up on a midterm.
The picture: a city of homes and roads#
Think of the Internet as a city. The homes are the computers, phones and servers people actually use. The roads and intersections in between are what move things from one home to another.
Networking splits the city into two parts, and those two words come up on every exam:
- The network edge is the homes, plus the short driveway that joins each home to the nearest road. In network terms: the devices people use, and the link that connects them to their provider.
- The network core is the mesh of roads and intersections in the middle. In network terms: the routers and the links that join them.
A device at the edge is called a host or an end system. Both words mean the same thing: a machine that runs applications, like your laptop or a web server. A router is a machine in the core whose only job is to pass data along toward where it is going.
A server is a host, not part of the core. It sits at the edge, even if it lives in a huge data center. The core is only the routers and links that carry data between hosts.
Vocabulary, before anything uses it#
| Term | What it means |
|---|---|
| Host (end system) | A device at the edge that runs applications: a laptop, phone, or server. |
| Router | A device in the core that receives packets and sends each one onward toward its destination. |
| Link | A connection between two devices. Its speed is its transmission rate, measured in bits per second (bps). |
| Access network | The network that joins a host to its first router, called the edge router. |
| Packet | A small chunk of a larger message, with an address on the front. |
| Packet switching | Moving data as packets that share links with everyone else, first come, first served. |
| Circuit switching | Reserving a fixed slice of every link along the path before any data is sent. |
| Store-and-forward | A router must receive a whole packet before it can start sending any of it onward. |
| Queue (buffer) | The waiting line of packets at a router when the outgoing link is busy. |
| FDM | Frequency-division multiplexing: each circuit gets its own band of frequencies, all the time. |
| TDM | Time-division multiplexing: each circuit gets the whole link, but only during its own repeating time slot. |
| ISP | Internet service provider: a company that runs part of the network and sells access to it. |
The edge: how hosts get connected#
Before data reaches the core, it has to leave your device. The access network is that first stretch. The common kinds:
| Access type | How it works, in one line |
|---|---|
| DSL | Uses the existing telephone line to the home. Voice, upload and download share the one wire by using different frequencies, so the phone and the Internet work at the same time. |
| Cable | Uses the cable TV network. The line from the street is shared by many homes, so speeds drop when the neighbors are busy. |
| Fiber to the home (FTTH) | Runs optical fiber all the way to the house. Very fast. |
| Ethernet | Wired connections inside companies and universities, usually to a switch in the building. |
| Wi-Fi | Wireless access over a short range, to an access point that is wired to the network. |
| Cellular (4G, 5G) | Wireless access over kilometers, through a mobile provider's base stations. |
Every link is built from some physical medium, the stuff the signal actually travels through. Media come in two families. Guided media keep the signal inside a solid path: twisted-pair copper wire, coaxial cable, and fiber optic cable. Unguided media let the signal spread through the air: radio, including Wi-Fi, cellular and satellite.
Fiber carries light, so it is immune to electrical interference and has very low signal loss over long distances. That is why the long-haul backbone of the Internet is fiber.
Packet switching and store-and-forward#
The Internet core uses packet switching. A long message is cut into packets. Each packet travels through a series of links and routers, sharing each link with everybody else's packets.
Routers work by store-and-forward. Picture a small ferry that will not leave the dock until every car in the group has boarded. A router is the same: it must receive the entire packet before it can send the first bit of it onto the next link.
Pushing a packet of L bits onto a link of rate R bits per second takes L/R seconds. So one hop costs L/R, and a path of N links costs N times that:
Store-and-forward delay over N links = N × L / R
(This ignores the time the signal spends travelling and any waiting in line. Those delays get their own guide.)
Queues and loss
Sometimes packets arrive at a router faster than its outgoing link can send them. The extra packets wait in a queue, also called a buffer. That waiting time is queuing delay. The buffer is a fixed size. If it is full when a packet arrives, the router drops that packet, and it is lost.
Forwarding vs routing
A router does two different jobs, and exams like to ask which is which:
| Job | Scope | What it does |
|---|---|---|
| Forwarding | Local, fast, per packet | Move an arriving packet from the input link to the right output link, by looking up its destination address in a table. |
| Routing | Network-wide, slower | Work out the paths packets should take, and fill in those tables. Routing algorithms do this. |
Forwarding uses the table. Routing builds the table. If a question says “move a packet from an input port to an output port”, the answer is forwarding.
Circuit switching: reserved lanes, FDM and TDM#
The old telephone network worked differently. Before a call starts, the network reserves a slice of capacity on every link along the path, and holds it for the whole call. That reserved path is a circuit. It is like booking a private lane on every road from your house to your friend's house, for as long as the visit lasts.
The good part: guaranteed speed, and no waiting in line. The bad part: the lane is yours even when you are silent, so nobody else can use it.
A link is split into circuits in one of two ways:
| Method | Full name | How the link is shared |
|---|---|---|
| FDM | Frequency-division multiplexing | Each circuit gets its own band of frequencies, all of the time. Like radio stations, each on its own channel. |
| TDM | Time-division multiplexing | Each circuit gets the whole bandwidth, but only during its own time slot, which repeats in every frame. |
Either way, if a link of rate R is split into n circuits, each circuit gets R/n.
FDM splits the frequencies. TDM splits the time. In FDM you always have a narrow channel. In TDM you periodically get the entire link for a short moment. The rate you end up with, R/n, is the same.
Worked example 1: file transfer over a circuit#
A 960,000-bit file is sent over a circuit-switched network. Every link uses TDM with 32 slots and has a rate of 2.048 Mbps. Setting up the circuit takes 250 ms. How long does it take to send the file, from the start of setup?
Step 1: Find the rate of one circuit. The link is shared by 32 slots, so one circuit gets a 32nd of it.
2,048,000 bps / 32 = 64,000 bps = 64 kbps
Step 2: Divide the file by that rate.
960,000 bits / 64,000 bps = 15 s
Step 3: Add the setup time.
15 s + 0.25 s = 15.25 s
Notice that the file never gets the full 2.048 Mbps, even if nobody else is using the link. The other 31 slots sit empty rather than help. That wasted capacity is the whole argument against circuit switching for bursty data.
The per-circuit rate worked out first, the division with units, and the setup time added on. Forgetting to divide by the number of slots gives 0.47 s plus setup, which is the most common wrong answer.
Worked example 2: how many users can share a link?#
Here is why the Internet chose packets. Most users are bursty: they send a lot for a moment, then nothing while they read.
A link runs at 2 Mbps. Each user needs 200 kbps while active, and is active only 10% of the time. How many users can circuit switching support? With packet switching and 40 users, how likely is it that too many are active at once?
Step 1: Circuit switching. Every user must be given a reserved 200 kbps, active or not.
2,000 kbps / 200 kbps = 10 users
Step 2: What “too many” means for packets. The link can carry 10 active users at full speed. Trouble only starts when 11 or more of the 40 are active at the same instant.
Step 3: Work out that chance. Each user is independently active with probability 0.1. The number active follows a binomial distribution with n = 40 and p = 0.1. Adding up the probabilities for 11 through 40 active users gives:
P(11 or more active out of 40) ≈ 0.0015
That is about 1.5 times in every 1,000 moments. So packet switching serves four times as many users on the same link, and almost never runs short. When it does, packets queue briefly rather than being refused.
Circuit switching reserves capacity whether you use it or not; packet switching shares it on demand, so it wins for bursty traffic and pays for that with queuing and the chance of loss.
The core at scale: a network of networks#
No single company runs the Internet. Hosts connect to an access ISP (your home provider or your university). Access ISPs cannot all connect directly to each other, since millions of links would be needed, so they connect upward in a hierarchy:
| Piece | Role |
|---|---|
| Access ISP | Connects homes, companies and campuses. Pays a larger ISP to reach the rest of the Internet. |
| Regional ISP | Collects many access ISPs in an area and connects them upward. |
| Tier-1 ISP | A very large global network that can reach every part of the Internet without paying anyone to carry its traffic. |
| IXP | Internet exchange point: a building where many networks meet and connect to each other directly. |
| Peering | Two networks at the same level connecting directly, usually without paying each other. |
| Content provider network | A private network run by a company like Google, reaching close to users so its traffic can skip upper tiers. |
The money flows one way: a customer pays its provider for carriage. Peering and IXPs exist to cut those bills, because traffic that crosses directly between two networks does not have to be paid for upstream.
An IXP is a meeting place, not a big ISP. It is where networks connect to each other. A tier-1 ISP is a network that carries traffic.
Practice#
Do the numbers on paper and write the units every time. Units are where most lost marks hide in this chapter.
- Name the two parts of the network, and say which one a web server in a data center belongs to.
- What does store-and-forward mean, and what does it cost per link?
- A 12,000-bit packet crosses 3 links, each 2 Mbps. Ignoring propagation and queuing, how long from the first bit leaving the source to the last bit arriving?
- Explain the difference between forwarding and routing in one sentence each.
- FDM or TDM? “Each user gets the full link bandwidth, but only for short repeating slots.”
- A 1,280,000-bit file is sent over a circuit on links of 1.536 Mbps using TDM with 24 slots. Setup takes 500 ms. What is the total time?
- A 1 Mbps link. Each user needs 100 kbps when active. How many users can circuit switching support, and why can packet switching support more?
- What happens to a packet that arrives at a router whose buffer is full?
- What is an IXP, and why do ISPs want to use one?
Solutions#
Find the first line where your working split from the solution, and redo the problem from that line. Do not just read the answer and nod.
- Edge and core. The edge is the hosts and their access links; the core is the routers and links in between. A server is a host, so it belongs to the edge, however large the data center.
- Whole packet first. A router must receive the entire packet before sending any of it onward. Each link costs L/R seconds, where L is the packet size in bits and R is the link rate.
- 18 ms. One link: 12,000 / 2,000,000 = 0.006 s = 6 ms. Three links, store-and-forward: 3 × 6 = 18 ms.
- Local vs network-wide. Forwarding moves one arriving packet to the correct output link using the router's table. Routing works out the paths across the network and builds those tables.
- TDM. Time-division gives each circuit all of the bandwidth during its own time slot. FDM would give each circuit a narrow frequency band all of the time.
- 20.5 s. Per circuit: 1,536,000 / 24 = 64,000 bps. Transfer: 1,280,000 / 64,000 = 20 s. Add setup: 20 + 0.5 = 20.5 s.
- 10 users; bursty traffic. 1,000 / 100 = 10 reserved circuits. Packet switching shares capacity only when users are actually sending. Because each user is active only part of the time, many more users rarely need the link all at once.
- It is dropped. The packet is lost. The buffer has a fixed size, and a full buffer has nowhere to put it. A higher layer, or the application, may send it again.
- A meeting point. An Internet exchange point is a place where many networks connect directly to each other. ISPs use it to peer, which avoids paying a larger provider to carry that traffic.
Test-readiness checklist#
- ☐ I can say what belongs to the edge and what belongs to the core.
- ☐ I can name at least four access network types and one fact about each.
- ☐ I can explain store-and-forward and compute N × L / R.
- ☐ I can tell forwarding from routing.
- ☐ I can explain FDM and TDM and compute a circuit's rate as R/n.
- ☐ I can compute a circuit-switched file transfer time, including setup.
- ☐ I can explain, with numbers, why packet switching supports more bursty users.
- ☐ I can describe access ISP, regional ISP, tier-1 ISP, IXP and peering.
Divide by the number of slots or bands before you divide the file size. Almost every wrong circuit-switching answer comes from using the full link rate for one circuit.
Frequently asked questions#
What is the difference between the network edge and the network core?
The edge is the hosts, meaning the devices that run applications, plus the access links that connect them. The core is the mesh of routers and links in the middle that carries data between hosts.
Why does the Internet use packet switching instead of circuit switching?
Because Internet traffic is bursty. Packet switching shares link capacity on demand, so the same link can serve many more users than reserved circuits could. The price is possible queuing delay and packet loss when too many send at once.
What is the difference between FDM and TDM?
FDM gives each circuit its own frequency band all of the time. TDM gives each circuit the entire link, but only during its own repeating time slot. Both give each of n circuits a rate of R/n.
How do you calculate file transfer time over a circuit-switched network?
Divide the link rate by the number of slots or bands to get one circuit's rate, divide the file size in bits by that rate, then add the circuit setup time.
What is store-and-forward transmission?
A router must receive an entire packet before it begins sending it onto the next link. Each link therefore adds L/R seconds, where L is the packet length in bits and R is the link rate.
Something here not clear? A topic you wish we covered? Tell us. We read every message, and a request is the fastest way to get a guide written.
Test yourself on this topic
The free networking practice app has 19 questions on this topic and 500 across the whole course, each with an explanation for every answer. Offline, no ads, no account.
Practise this topic →