Understand why networks work, then trace what happens when your software sends a request.
Start with the OSI modelNetworking is the system that lets software communicate across machines. Learn the big picture first, then connect protocols, routing, reliability, wireless, security, and media delivery to the decisions developers make.
Use this map to ask a practical question during design or debugging: which job is failing, and which component owns it?
At each step, notice what this layer adds or checks; the receiver removes or interprets those headers in reverse order.
Start at the application layer. Select any layer above to inspect its job.
How to use the model: OSI is a teaching and troubleshooting map, not a promise that every Internet protocol fits one box. The Internet stack commonly groups OSI layers 5–7 as application, layer 4 as transport, layer 3 as Internet, and layers 1–2 as link. TLS, QUIC, and session behavior can cross these boundaries.
Work through nine course modules, each built around the networking knowledge software engineers use in real products. Complete all 41 guided lessons and demonstrate each module’s hands-on lab to earn your certificate of completion.
Start with the whole path: applications create data, networks carry packets, and protocols coordinate each step.
Applications run on end systems at the edge. Access networks connect them to routers; the core forwards packets across interconnected networks.
The Internet is a network of networks. End systems connect through access networks to routers, and providers interconnect those routers; protocols let independent networks exchange data.
In practice: A page request can cross home Wi-Fi, an ISP, transit networks, and a server network.
Packet switching divides messages and shares link capacity as packets arrive. Circuit switching reserves a path’s capacity in advance.
Packet switching divides messages into packets that share link capacity. Routers store packets, inspect headers, then forward them; circuit switching instead reserves capacity along a path.
In practice: Video and web traffic can share a link, but may queue when arrivals exceed its transmission rate.
Delay includes processing, queueing, transmission, and propagation. End-to-end throughput is limited by the bottleneck link.
dtrans = L / RNodal delay combines processing, queueing, transmission, and propagation. Transmission is packet length divided by link rate; propagation is distance divided by signal speed. Queueing varies with traffic.
In practice: On a slow access link, a large packet may be transmission-limited; across a long fiber path, propagation may dominate.
Each layer provides a service to the one above. Headers are added as data moves down the stack, then removed at the receiver. The chapter closes with network attacks and Internet history.
Layers divide networking into services and protocols. Encapsulation adds headers as an application message becomes a segment, datagram, and frame; the receiver removes them in reverse order.
In practice: A browser request keeps its meaning while TCP, IP, and Ethernet add delivery information.
Key distinction: Transmission delay is the time to place packet bits onto a link; propagation delay is the time those bits travel across it.
Adjust a slider to see how packet size, link rate, distance, and queues shape delay.
Applications define the messages exchanged by processes. Transport carries those messages between hosts.
Client-server applications contact always-on servers. Peer-to-peer applications let hosts act as both clients and servers. Sockets connect an application process to transport.
Application architecture describes which processes communicate and where they run. Client-server systems use an often-available server; peer-to-peer systems let hosts request and provide service. A socket connects an app to transport.
In practice: A mobile app opens a socket to an API; a file-sharing app can connect to several peers.
A browser sends an HTTP request; a server returns a response. HTTP is stateless, while cookies and server-side state can support sessions.
HTTP is request-response: a client asks for a resource and a server returns a status, headers, and possibly a body. Persistent connections reuse TCP; caches avoid round trips; cookies carry state.
In practice: A browser can reuse one connection for several assets and load a repeat image from cache.
DNS MX records locate receiving mail servers; SMTP relays and queues messages; IMAP or POP3 gives recipients mailbox access.
Email uses DNS for destination discovery and SMTP for submission and server-to-server relay. Receiving servers can accept messages into durable queues; mailbox access happens afterward through a separate service.
In practice: Inspect MX records, SMTP response codes, queue age, and mailbox filtering as separate checkpoints.
Peer-to-peer systems distribute work across hosts. Video services use HTTP streaming and CDNs to place content closer to viewers. Applications use UDP or TCP sockets.
DNS is a distributed directory mapping names to records. Peer-to-peer apps distribute work among hosts; video services use adaptive HTTP streaming and CDNs to put content near viewers.
In practice: DNS resolves a site’s name, then a CDN can direct the browser to a nearby cache.
Connection to Chapter 3: HTTP, email, and streaming define application messages; transport protocols determine how those messages move between processes.
Each additional non-persistent object needs a new TCP connection and a request/response round trip.
Transport uses ports to deliver data to the right process. TCP adds reliability, flow control, and congestion control; UDP keeps the service lightweight.
Multiplexing gathers data from application sockets. Demultiplexing uses port numbers to pass each segment to the correct process.
Transport multiplexes data from application sockets and demultiplexes segments to the right process. Port numbers identify endpoints; TCP also uses connection state, while UDP uses a simpler endpoint mapping.
In practice: A laptop can run a browser, call, and DNS client at once because the OS separates their traffic.
UDP sends independent segments without a handshake or built-in retransmission. Its checksum helps detect errors in a segment.
UDP provides ports and a checksum without a handshake, retransmission, or built-in congestion control. It fits applications that need low overhead and can manage timing or loss themselves.
In practice: A real-time app may skip a late audio packet instead of retransmitting it after it is no longer useful.
Sequence numbers, acknowledgments, timers, and retransmissions let TCP provide reliable, in-order delivery over IP.
Reliable transfer uses sequence numbers, acknowledgments, checksums, and timers to detect loss and restore in-order delivery. Go-Back-N repeats a missing packet and later unacknowledged packets; Selective Repeat repeats only missing packets and buffers later arrivals.
In practice: A sending window keeps multiple packets in flight while tracking what the receiver confirmed.
Flow control keeps a fast sender from overwhelming the receiver. Congestion control responds to load inside the network.
Flow control protects a receiver’s buffer; congestion control responds to load inside the network. TCP estimates round-trip time for timeouts and adjusts its sending rate when it detects congestion.
In practice: A sender may be limited by a slow receiver, a congested path, or both; these controls solve different problems.
Reliable transfer: Go-Back-N retransmits the missing packet and later unacknowledged packets. Selective Repeat retransmits only missing packets and buffers out-of-order arrivals.
Go-Back-N discards packets received beyond a gap. A timeout makes the sender repeat every unacknowledged packet in its window.
Routing plans paths across the network. Forwarding moves each arriving packet from a router’s input to its chosen output.
The control plane determines routes. The data plane uses a local forwarding table to move each packet to an output link.
Routing computes network-wide paths; forwarding applies a router’s local table to each arriving datagram. The control plane installs forwarding state, while the data plane handles packets at line speed.
In practice: A route change updates forwarding tables; subsequent packets follow the selected next hop.
An input port examines the packet header; the switching fabric carries it across the router; an output port queues and transmits it.
An input port receives a frame, extracts its destination, and looks up a forwarding entry. The switching fabric moves the packet to an output port, where buffering and scheduling determine when it leaves.
In practice: If two inputs compete for one output, packets may queue even when their routes are correct.
IPv4 and IPv6 datagrams carry addresses. Subnets and longest-prefix matching guide forwarding; NAT translates addresses at network boundaries.
IPv4 uses 32-bit addresses; subnet prefixes divide network and host parts. NAT maps private addresses to public address space; IPv6 expands addresses. Longest-prefix matching chooses the most specific matching entry.
In practice: A home router translates private device addresses; an ISP can prefer a /24 route over a default.
Generalized forwarding can match on packet fields and apply actions such as forwarding, dropping, or modifying the packet.
Generalized forwarding matches packet fields and applies actions such as forwarding, dropping, or rewriting. SDN separates controller logic from devices that execute rules; scheduling manages competing queues.
In practice: A controller can install a traffic rule while a router scheduler chooses which queued packet transmits next.
Queueing matters: If packets arrive at an output faster than its link can send them, they wait in a queue; when the buffer fills, packets are dropped. Scheduling decides which waiting packet goes next.
| Network prefix | Next hop |
|---|---|
| 0.0.0.0/0 | Link 4 · default |
| 203.0.113.0/24 | Link 1 |
| 203.0.113.128/25 | Link 2 |
| 203.0.113.144/28 | Link 3 |
How routers find paths and coordinate across networks. Open any module for a five-page lesson with an explanation, a worked example, guided practice, and a checkpoint.
Routers build a view of network links, then run Dijkstra’s algorithm to find least-cost paths.
Link-state routers learn the network graph and link costs, then run Dijkstra’s algorithm to find least-cost paths. Link-state advertisements keep topology views aligned; each router computes its own forwarding choices.
In practice: When a link cost changes, routers advertise the update and recompute paths.
Routers exchange distance estimates with neighbors and update paths using the Bellman–Ford relation.
Distance-vector routers exchange destination costs with neighbors and apply Bellman–Ford: route cost equals link cost to a neighbor plus that neighbor’s advertised cost. Updates can converge slowly after failures.
In practice: A neighbor may offer a cheaper route, but bad news can spread slowly and create count-to-infinity behavior.
OSPF distributes link-state information inside an autonomous system; areas limit the scope of updates.
OSPF is an intra-autonomous-system link-state protocol. Routers flood advertisements, build a shared topology database, and compute paths; areas and designated routers help limit overhead.
In practice: A campus can keep detailed routes inside areas and use summaries between them.
BGP advertises reachability between autonomous systems, combining policy choices with path information.
BGP exchanges reachability between autonomous systems using path-vector advertisements. Operators choose routes using attributes and policy, not just hop count; AS-PATH helps prevent loops and NEXT-HOP identifies the next router.
In practice: Providers may choose different paths because of business relationships or traffic-engineering goals.
A logically centralized controller can compute and distribute forwarding rules to network devices.
An SDN controller maintains a network view and programs forwarding devices through a protocol such as OpenFlow. Control applications can implement routing, access policy, or traffic engineering; devices still forward packets.
In practice: A controller can coordinate changes across several switches after one policy update.
ICMP reports network-layer errors and diagnostics; SNMP provides a framework for monitoring managed devices.
ICMP carries network-layer errors and diagnostics, supporting ping and traceroute. SNMP uses a manager-agent model to read device counters and configuration; management data helps operators assess network health.
In practice: Interface error counters can reveal a bad link; ICMP time-exceeded messages help trace a route.
How neighboring devices deliver frames over a link. Open any module for a five-page lesson with an explanation, a worked example, guided practice, and a checkpoint.
Link protocols frame datagrams and detect corruption with tools such as checksums, parity, and CRC.
The link layer moves a datagram across one link between adjacent nodes. Its services may include framing, medium access, reliable delivery, and error detection; the exact set depends on the link technology.
In practice: Ethernet and Wi-Fi both carry IP datagrams, but their framing and access rules differ.
Parity, checksums, and cyclic redundancy checks detect corrupted frames; recovery requires retransmission or additional redundancy.
Parity, checksums, and cyclic redundancy checks let a receiver detect corrupted frames. CRCs detect many common burst errors, but detecting corruption does not always let a receiver repair the data.
In practice: An Ethernet CRC can expose corruption so the frame is discarded and a higher layer may retransmit.
Channel partitioning, random access, and taking-turns protocols coordinate devices that share a broadcast medium.
Multiple-access protocols coordinate devices sharing a broadcast medium. Channel partitioning reserves frequencies or turns; random access allows contention; taking-turns protocols grant transmission opportunities in sequence.
In practice: Wi-Fi uses contention to share airtime, while cable networks coordinate upstream transmissions.
ARP resolves a local IPv4 next hop to a MAC address; Ethernet switches learn source addresses and forward frames across a LAN.
A link-layer address identifies an interface on a local link. ARP resolves an IPv4 address to a MAC address; Ethernet switches learn source addresses and forward frames toward the right port.
In practice: For an off-subnet destination, a laptop sends the frame to its router’s MAC address, not the remote server’s.
VLANs segment switched LANs, MPLS steers provider traffic with labels, and data-center fabrics connect large server populations.
VLANs create logical LAN segments across physical switches; MPLS uses labels to steer packets through provider networks. Data centers use fabrics with many paths to connect servers at scale.
In practice: A VLAN can isolate guest devices; a data-center fabric provides parallel paths between racks.
How networks handle radio links and moving devices. Open any module for a five-page lesson with an explanation, a worked example, guided practice, and a checkpoint.
Radio signals face attenuation, interference, and multipath; wireless links adapt to changing conditions.
Wireless links share spectrum and face attenuation, interference, and multipath fading. Signal quality changes with distance and obstacles; protocols adapt rates and use error control.
In practice: Moving a laptop behind a wall can reduce signal quality even when its IP address is unchanged.
802.11 uses association and channel access rules to connect stations through an access point.
802.11 networks use stations, access points, and a distribution system. Wi-Fi stations contend for the channel and use acknowledgments because wireless collisions are difficult to detect while sending.
In practice: A busy access point can increase delay as connected devices compete for airtime.
Wireless LANs and personal-area networks use different link rules; Bluetooth and Zigbee trade range, rate, and power for nearby devices.
Wi-Fi frames carry addresses for wireless delivery and bridging; advanced 802.11 features improve throughput. Bluetooth and Zigbee support personal-area uses with different range, power, and rate tradeoffs.
In practice: A wearable may use Bluetooth to reach a phone, which forwards data over Wi-Fi or cellular.
Cellular radio access, carrier-core services, and handoff keep a device connected as its point of attachment changes.
Cellular networks combine radio access nodes with a core that authenticates devices, assigns service, and connects traffic to the Internet. LTE organizes communication around base stations and packet-core functions.
In practice: The radio link ends at a base station; the carrier core manages the device’s session and traffic.
How communication is protected and attacks are detected. Open any module for a five-page lesson with an explanation, a worked example, guided practice, and a checkpoint.
Threat models identify what an attacker can observe, modify, impersonate, or disrupt.
Security starts with a threat model: identify assets, attackers, and needed guarantees. Common goals include confidentiality, integrity, authentication, and availability; each mechanism addresses particular threats.
In practice: A login API may need to hide passwords in transit, verify the server, and limit repeated guesses.
Symmetric and public-key cryptography support confidentiality, message integrity, and digital signatures.
Symmetric cryptography uses a shared secret for efficient encryption; public-key cryptography uses a key pair to establish secrets or prove identity. Protocols combine them because public-key operations cost more for bulk data.
In practice: TLS can use public-key techniques during setup, then protect session data with symmetric encryption.
Authentication verifies identity; signatures bind a message to a signer and help detect changes.
A cryptographic hash produces a digest; a message authentication code uses a secret to detect tampering and authenticate a sender. Digital signatures use a private key to sign and a public key to verify.
In practice: A signed software release lets users check who created it and whether the file changed.
Fresh challenges and certificate validation help a client prove which server it is talking to and resist replay.
Endpoint authentication must distinguish a real participant from an impostor, even when messages can be replayed. Nonces and challenge-response exchanges make old authentication messages unusable in a new session.
In practice: A fresh random challenge prevents a captured response from simply being replayed later.
Secure email, TLS, IPsec/VPNs, wireless protection, firewalls, and intrusion detection protect different trust boundaries.
Secure email combines encryption for confidentiality with signatures or message authentication for integrity and sender verification. PGP packages these tools around user-managed keys; key distribution and trust remain part of the problem.
In practice: Encryption hides message contents from intermediaries; a signature helps verify who sent it.
How networks deliver audio and video under timing constraints. Open any module for a five-page lesson with an explanation, a worked example, guided practice, and a checkpoint.
Audio and video traffic has timing and rate needs that differ from ordinary file transfers.
Video has high, variable bit rates; audio is continuous and timing-sensitive. Multimedia applications include stored streaming, live streaming, and conversational services, each with different delay and loss tolerance.
In practice: A movie can buffer ahead; a voice call cannot wait seconds to refill a buffer.
Adaptive streaming selects chunk quality based on bandwidth, buffer state, and playback conditions.
Stored video can stream over UDP or HTTP. HTTP adaptive streaming divides media into chunks encoded at different quality levels; the client chooses each next chunk based on network conditions and buffer state.
In practice: If bandwidth drops, a player can request a lower-bitrate chunk to avoid a playback stall.
Real-time voice systems trade a small amount of loss tolerance for low end-to-end delay.
Best-effort IP does not guarantee delay, bandwidth, or loss. VoIP uses packetization, playback buffers, and loss recovery to balance smooth audio against conversational delay.
In practice: A larger jitter buffer smooths uneven arrivals but makes speakers hear one another later.
RTP carries real-time media; SIP helps establish, manage, and end multimedia sessions.
RTP carries real-time media with sequence numbers and timestamps so a receiver can reconstruct timing; it does not guarantee delivery. SIP establishes, manages, and ends multimedia sessions.
In practice: A voice service can use SIP to set up a call and RTP to carry audio.
QoS tools such as scheduling, resource reservation, and Diffserv classes can manage competing traffic.
Networks support multimedia by dimensioning capacity, prioritizing traffic classes, or reserving resources. Diffserv marks packets for per-hop treatment; admission control can reject flows when resources are insufficient.
In practice: A router may give voice low-delay service while limiting that queue so other traffic still gets capacity.
Use the evidence to locate the first failing boundary, then choose a testable cause. This short practice lab is optional and does not change certificate requirements.
Finish the 41 guided lessons, then demonstrate one hands-on lab in each of the nine course modules. Your progress is saved in this browser.
Complete the lessons and demonstrate each module lab to unlock your certificate.
CERTIFICATE OF COMPLETION
has completed the Nine-Module Interactive Course in Computer Networking, including all 41 guided lessons and nine demonstrated hands-on labs.
Completed