What Is a Network?

What a network is

A network is a group of devices connected so that they can exchange data. The reason networks exist is sharing: sharing information (email, files, a web page, a video call) and sharing resources (one printer for a whole floor, one internet connection for a household, one storage system for a data center). Without a network, every computer would be an island, and moving data would mean carrying it around on a disk.

Every network, from a home to the internet, has the same parts:

• Nodes: the devices on the network. End devices (hosts) are where data starts and ends: laptops, phones, servers, printers, IP phones, cameras. Intermediary devices move data between them: switches, routers, wireless access points and firewalls. See Network Devices.
• Links: the paths between nodes. A link can be a copper cable, a fiber-optic cable or a radio signal (Wi-Fi, cellular). See Cabling and the Physical Layer.
• Protocols: the agreed rules that both sides follow, so a message sent by one device can be understood by another. They are organized in layers; see The OSI and TCP/IP Models.

LAN, WAN and everything in between

Networks are grouped by the area they cover, which also says a lot about who owns them and what they cost.

Local area network (LAN)

A local area network (LAN) connects devices in a limited area — a home, an office floor, a building or a campus — and is owned and managed by one organization. LANs are fast (1 Gbit/s to the desk and 10 Gbit/s or more between switches is normal) and cheap per bit, because the organization owns the cables. Wired LANs use Ethernet; a wireless LAN (WLAN) uses Wi-Fi. Both are usually part of the same network.

Wide area network (WAN)

A wide area network (WAN) connects LANs that are far apart: a head office with its branch offices, or an office with a data center in another city. An organization rarely owns the long-distance cables; it rents connections from a service provider (a leased line, an MPLS (Multiprotocol Label Switching) service, broadband internet with a virtual private network (VPN) on top). That is why WAN links are slower and more expensive per bit than LAN links, and why a WAN outage usually means a ticket with the provider.

MAN, campus and other names

You will meet a few more names. They describe scale, not technology:

• Campus area network (CAN): several buildings of one organization close together, such as a university or a hospital, linked by its own fiber.
• Metropolitan area network (MAN): a network across a city, often run by a city or a provider.
• Personal area network (PAN): the few meters around one person, such as Bluetooth earbuds and a phone.
• Storage area network (SAN): a dedicated network that connects servers to shared storage in a data center.

The internet: a network of networks

The internet is not one network. It is tens of thousands of independently run networks — internet service providers (ISPs), cloud providers, universities, companies — that agree to exchange traffic with each other. Each one routes traffic for its own addresses and hands the rest to a neighbor. That the whole system works without a central owner is the result of shared, open protocols such as the Internet Protocol (IP). Your home or office LAN is one small network at the edge of it.

Topologies

A topology is the layout of a network: which nodes are connected to which. It decides what happens when a link or a device fails, and how much cabling you need.

Physical versus logical topology

The physical topology is how the cables and devices are actually arranged. The logical topology is how data flows. They can differ: an old Ethernet hub was a physical star, but because it repeated every signal to every port it behaved logically like a bus. Token Ring was the reverse case: stations were cabled in a physical star to a central unit, but the token passed from station to station in a logical ring. Today a typical office is a physical star of desks around a switch per floor, those switches connect in a star or partial mesh to a core, and the whole thing is a hybrid. Data centers often use a spine-leaf design, a partial mesh in which every access (leaf) switch connects to every core (spine) switch.

Client-server and peer-to-peer

Topology describes cables; these two models describe roles: who offers a service and who uses it.

In the client-server model, a server offers a service (web pages, email, files, name resolution) and many clients send it requests. The server is managed centrally, which makes backups, security and access control easier. Almost everything in a company network works this way: a browser is a client of a web server, a laptop is a client of the Dynamic Host Configuration Protocol (DHCP) and Domain Name System (DNS) servers.

In a peer-to-peer (P2P) network, every device is both client and server: peers share their own files or resources directly with each other. A small office where two PCs share a folder is peer-to-peer, and so are file-sharing systems that spread downloads over thousands of peers. Peer-to-peer needs no dedicated server, but it does not scale well to managed environments: there is no central place for accounts, backups or security policy.

Note that "server" is a role, not a kind of hardware. A laptop that shares a folder is acting as a server for that moment.

Measuring performance

"The network is slow" is the most common complaint a network operations (NetOps) team gets. To act on it, you need words that separate how much from how fast. Think of a highway.

• Bandwidth is the capacity of a link: the maximum number of bits it can carry per second. It is the number of lanes and the speed limit of the highway. A 1 Gbit/s Ethernet port has a bandwidth of 1 Gbit/s, whatever is going on around it.
• Throughput is the amount of data that actually gets through per second, measured over a real transfer. It is how many cars really pass per hour. It is always lower than the bandwidth, because of protocol overhead (headers), other traffic sharing the link, retransmissions and the slowest link on the path.
• Latency is the delay: how long it takes data to get from one end to the other, usually measured in milliseconds as a round-trip time (RTT) by ping. It is how long one car takes to drive the route. Adding lanes does not make the route shorter, which is why more bandwidth does not fix latency.

Where latency comes from

Latency is the sum of four delays:

• Propagation delay: the time a signal travels along the cable. Light in fiber covers about 200 km per millisecond, so a 6,000 km path adds about 30 ms one way. Nothing can make this smaller except a shorter path.
• Serialization delay: the time to put all bits of a frame onto the wire. A 1,500-byte frame takes 12 µs at 1 Gbit/s and 1.2 ms at 10 Mbit/s.
• Queuing delay: time waiting in a device's buffer behind other traffic. This is the part that grows when a link is congested.
• Processing delay: the time a device needs to look at a header and decide where the data goes. Usually tiny.

Jitter and packet loss

Two more measurements show up on every monitoring dashboard. Jitter is the variation in latency from one packet to the next. A file download does not care, but a voice or video call does: packets that arrive at irregular intervals make speech choppy. Packet loss is the percentage of packets that never arrive, because a buffer overflowed, a cable is damaged or a wireless signal is weak. TCP resends lost data, which lowers throughput; real-time traffic simply loses it. Voice over IP (VoIP) is usually considered good below about 150 ms one-way latency, 30 ms jitter and 1% loss.

Units: bits, bytes and prefixes

Data rates are always counted in bits per second, written bps or bit/s, with decimal prefixes: 1 kbit/s = 1,000 bit/s, 1 Mbit/s (Mbps) = 1,000,000 bit/s, 1 Gbit/s (Gbps) = 10⁹ bit/s. File sizes, on the other hand, are counted in bytes (8 bits), and operating systems often use binary prefixes for them (1 MiB = 1,048,576 bytes). A lowercase b means bits, an uppercase B means bytes: 100 Mbps is 12.5 MB/s. For the bit and byte themselves, see Number Systems.

file_mb = 500                         # file size in megabytes (1 MB = 1,000,000 bytes)
file_bits = file_mb * 1_000_000 * 8   # 8 bits per byte

bandwidth = 100_000_000               # link: 100 Mbit/s
throughput = 80_000_000               # measured during the transfer: 80 Mbit/s

print(f"File size:           {file_bits:,} bits")
print(f"Link in MB/s:        {bandwidth / 8 / 1_000_000} MB/s")
print(f"Best case:           {file_bits / bandwidth:.0f} s")
print(f"Real transfer:       {file_bits / throughput:.0f} s")
print(f"Forgetting the x8:   {file_mb * 1_000_000 / bandwidth:.0f} s  (wrong)")

File size:           4,000,000,000 bits
Link in MB/s:        12.5 MB/s
Best case:           40 s
Real transfer:       50 s
Forgetting the x8:   5 s  (wrong)

What a network engineer does all day

A NetOps team keeps the network available, fast and secure. On a normal day that means:

• Tickets: "Wi-Fi drops in meeting room 3", "new employee needs a network port", "VPN is slow". Each one is narrowed down layer by layer; see Troubleshooting Tools and Method.
• Monitoring: dashboards that show which devices and links are up, how busy each link is (utilization), and error counters. Alerts arrive before users notice; see Operational Services: NTP, SNMP and Syslog.
• Changes: adding a VLAN, opening a firewall rule, upgrading switch software. Changes go through change management, are planned for a maintenance window and always have a way back.
• Documentation: network diagrams, IP address plans and cable records. When something breaks at 3 a.m., good documentation saves hours.
• Capacity planning and projects: noticing that a link is at 80% every afternoon, and upgrading it before it becomes a problem.
• Automation: more and more of this work is done with scripts and tools instead of by hand; see Network Automation.