Computer networks: foundations and end-to-end communication

1. Nodes, links, interfaces, and services

A node is a participating device: a host, router, switch, or other networked system. A link provides communication between interfaces. An interface is the attachment point to that link and has relevant configuration and addressing. A computer may have multiple interfaces, such as wired Ethernet and Wi-Fi, with different addresses and routes.

Hosts usually originate or consume application data. Intermediate devices forward traffic according to rules. Servers provide services; clients request them. These are roles, so one host can be a client of one service and a server for another. Peer-to-peer arrangements distribute responsibilities between peers rather than depending on one application server for every operation.

Networks provide resource sharing, communication, remote access, distributed computation, and access to services. Shared access creates dependencies and risks: a network outage can interrupt a service even when both endpoint computers still work. Authentication, authorisation, confidentiality, and availability must be designed explicitly.

2. Scope and ownership

PAN, LAN, MAN, and WAN describe common scopes: personal vicinity, a local site, a metropolitan area, and broader geographical connection. They are useful descriptive categories, not rigid universal distance limits. A campus network may combine several LANs over fibre, and a WAN may connect sites through a service provider.

A LAN is not necessarily private or secure merely because it is small. A WAN is not automatically the public Internet. Ownership, administration, technology, and geographical scope are separate properties. A private link can span countries; a public Wi-Fi LAN can expose local access to strangers.

An intranet supplies internal organisational network services. An extranet provides controlled access to selected external parties. The Internet is a global interconnection of networks using Internet protocols. The Web is a set of services and resources typically accessed through HTTP-based applications over that infrastructure. Email, remote shells, and many other applications also use Internet networks.

3. Circuit, packet, and layered thinking

Most Internet communication sends data in packets. A packet carries headers used for delivery and a payload containing information from a higher layer. A large application message may be divided across many packets; packet boundaries need not match application message boundaries.

Layers divide responsibilities. The link layer handles communication on a particular link; the Internet layer addresses and routes packets across networks; the transport layer provides endpoint communication services; applications define meaning and operations. Encapsulation wraps higher-layer data inside lower-layer units. The receiver interprets the wrappers in the reverse direction.

Layering allows one application to work over different link technologies. It is a model, not a promise that every device performs exactly one layer’s work. Firewalls, proxies, and offload hardware can examine or implement several responsibilities.

Application data → transport unit → IP packet → link frame → physical signal
At the receiver: signal → frame → IP packet → transport data → application

4. Addresses answer different delivery questions

A link-layer address identifies an interface for delivery on a link. An IP address identifies an interface or endpoint in the Internet-layer addressing scheme. A port identifies a transport service or communication endpoint within a host’s transport context. A DNS name is an application-friendly identifier whose records may supply addresses. These are different namespaces, not four interchangeable names for a person.

For an off-link destination, a host sends the link frame toward its next-hop router while the IP packet identifies the remote destination. Routers choose subsequent next hops. The Ethernet destination on the first link is therefore usually the router’s interface, not the remote web server’s Ethernet address.

An IPv4 prefix describes an address range. With prefix /24, 24 of 32 bits identify the prefix and 8 remain for addresses within it: 2^8 = 256 addresses. Conventional /24 subnets reserve network and directed-broadcast addresses, commonly leaving 254 host addresses. This convention must not be generalised to every prefix or to IPv6.

5. A request through a network

Imagine a laptop requesting a campus service on a different IP subnet. It needs a valid local configuration and route. It may resolve the service name through DNS, determine the next hop, resolve the next-hop link address, and send frames carrying transport and IP data. The switch forwards on the local network; a router forwards toward the destination network. The application server receives the request and constructs a response.

For HTTPS, secure transport is established and the server identity is checked using the relevant certificate and trust rules. The response follows routing decisions that need not be identical to the request’s path. Caches, proxies, NAT, and load balancers can alter the practical flow. This simplified trace supplies a starting model, not a claim that every web request repeats every setup step.

If name resolution fails, that differs from IP reachability failing. If the server returns an application error, packets may have been delivered correctly. Diagnose the failed responsibility rather than describing every problem as “the Internet is down”.

6. Performance, resilience, and network effects

Bandwidth is a capacity measure, typically bits per second. Throughput is the observed delivery rate. Goodput counts useful application data rather than all transmitted overhead. Latency is time taken for an operation or delivery; round-trip time includes a path out and back. Low latency and high throughput are distinct goals.

The narrowest capacity along a path can limit steady throughput, but congestion, loss, protocol behaviour, endpoints, and storage can reduce it further. A queue adds delay when arrivals exceed service temporarily. Buffers postpone loss but do not create capacity; excessively large queues can create long interactive delays.

Reliability requires considering independent failures and recovery. A redundant link does not protect against a shared failed switch or power supply. Network security needs access boundaries, secure protocols, updates, monitoring, and recovery. Connectivity alone provides none of these guarantees.

Worked example

A 100 MB decimal file contains 800 million bits. On a 20 Mb/s bottleneck, ideal payload transfer time is 800/20 = 40 seconds. Headers, retransmissions, setup, and application delays increase the real time. If the path RTT is 80 ms, bandwidth-delay product is 20,000,000 × 0.08 = 1,600,000 bits, or 200,000 bytes. This estimates data needed in flight to use that rate under the simplified path model; it is not a guarantee of actual throughput.