IB Computer Science · Theme A: Networks

A2.2 — Network Architecture

Everything the syllabus asks for on topologies, servers, networking models and segmentation — with the factor tables examiners actually award marks for, and practice questions with model answers.

SL + HL shared content A2.2.2 servers — HL only Aligned to the 2027 guide Mr Tzanis · IB CS Yr 2

Start here How to use this guide

Three passes beats one long read. Each section below is written to be revised in about 10 minutes.

Pass 1 — Learn

Read sections 1–4 in order. For each, say the factor list out loud without looking.

  • Topologies: 5 factors
  • Servers: 5 types × 4 factors
  • Models: benefits vs drawbacks
  • Segmentation: 3 mechanisms

Pass 2 — Test

Do the practice questions in section 6 closed book. Then mark yourself against the model answers and note every point you missed.

Pass 3 — Recall

Only the glossary and the traps. Rewrite the five factor tables from memory on blank paper — that is exactly what a 6-mark question rewards.

Exam reality check

A2.2 is an application topic, not a definitions topic. Nearly every question gives you a scenario (a school, a small office, a stadium) and then asks you to apply the factor list to it. A generic answer that never mentions the scenario caps out around half marks.

A2.2.1 Network topologies

Syllabus: describe the functions and practical applications of network topologies (star, mesh, hybrid), considering reliability, transmission speed, scalability, data collisions and cost.

SW
Star — every device has its own link to a central switch
Full mesh — n(n−1)/2 links; every node reaches every other directly
Hybrid — meshed (or partially meshed) core with star branches

What each one actually does

  • Star. All hosts connect to one central device (switch, or an access point in Wi-Fi). Traffic between two hosts passes through the centre. Functionally it creates one central point of control and, with a switch, a dedicated link per host.
  • Mesh. Nodes interconnect — full mesh connects every pair, partial mesh connects each node to several others. Traffic can be routed around a failed link, so the topology is self-healing.
  • Hybrid. Any combination, usually deliberately: a reliable meshed backbone where traffic volume and uptime matter, with cheap star "leaves" hanging off it where they do not.

The five factors — the table you must be able to write

FactorStarMesh (esp. full)Hybrid
Reliability Weak point: if the central switch fails, the whole segment is down. A single cable failure only loses one host, though. Strongest: no single point of failure — traffic re-routes around a broken link or dead node. High where it is meshed, star-level where it is not — chosen per branch.
Transmission speed Fast and predictable: each host gets a dedicated full-duplex link, so speed depends on the switch's capacity rather than on other hosts. Variable. Long routes mean more hops, each adding latency; wireless mesh shares one radio channel, so throughput falls as nodes are added. Fast at the edge (dedicated star links), with a fast core; latency rises with hop count across the mesh.
Scalability Easy: add a host with one cable / one port. Scales until the switch runs out of ports, then add a switch (i.e. go hybrid). Poor for full mesh: links grow as n(n−1)/2, so adding one node needs a new link to every existing node. Partial mesh scales better. Good: scale out at the star edge without touching the core.
Data collisions Low with a switch (each link is its own collision domain; full duplex removes collisions entirely). Higher with a hub, where all hosts share one domain. Low on wired point-to-point links, but high on wireless mesh — all nodes contend for the same channel, so collisions and retransmissions rise. Low at the star edge; depends on the core's switching/routing.
Cost Cheapest: one switch, one cable run per host, minimal configuration. Most expensive: cabling and an interface per link, plus complex configuration and management. Middle: cost is spent only where reliability is needed.

Practical applications to quote in answers

  • Star — home networks and small offices (one router/switch in the middle), school IT labs, and in practice any switched Ethernet LAN. Also the logical shape of Wi-Fi, with the access point at the centre.
  • Mesh — WAN and internet backbones, ISP and military networks, data-centre fabrics, city-wide wireless mesh, and home "mesh Wi-Fi" kits where nodes relay to each other to cover a large house.
  • Hybrid — a corporate campus or university: meshed core between buildings or floors, star to each desk; a data centre with a redundant spine and star-leaf switches.
Answer template — "which topology and why"

Factor → direction → because → scenario link. e.g. "Reliability: in a full mesh there is no single point of failure, because each node has at least two independent paths and traffic re-routes automatically — which matters for the hospital's patient-record system, where a switch failure cannot take the ward offline."

A2.2.2 · HL only Servers and their functions

Syllabus: describe the function of servers — DNS, DHCP, file, mail, proxy, web — considering function, scalability, reliability and security.

ServerFunctionScalabilityReliabilitySecurity
DNS Resolves a human-readable domain name to an IP address by querying the DNS hierarchy; caches results for the TTL. Distributed by design — the hierarchy splits worldwide load; local caching resolvers cut repeat queries; anycast spreads traffic. Redundant name servers per zone, caching (so an outage is partly absorbed) and short TTLs for failover. DNSSEC signs records to stop spoofing; guards against cache poisoning and DNS-amplification DDoS.
DHCP Automatically assigns IP configuration (address, subnet mask, gateway, DNS) to hosts for a lease period — the DORA exchange. One server configures thousands of hosts; scopes, reservations and multiple servers per subnet let the network grow without manual work. Leases renew automatically; a pair of DHCP servers (failover) avoids a single point of failure for joining the network. Rogue DHCP servers can hand out a malicious gateway; DHCP snooping on switches blocks untrusted servers.
File Central storage of shared files with permissions, quotas and scheduled backups; clients mount shares over SMB/NFS. Add disks or nodes; RAID and NAS/SAN arrays scale capacity and throughput; quotas stop one user consuming everything. RAID survives disk failure; backups and snapshots allow restore; redundant power and network links. Per-user ACLs and groups, encryption at rest and in transit, audit logging of access.
Mail Stores and forwards email: SMTP to send/relay between servers, IMAP or POP3 for clients to retrieve; filters spam and malware. Queues and multiple MX records spread inbound load; storage tiers and archiving cope with mailbox growth. Spooling means a message waits and retries if the destination is down; backup MX records accept mail when the primary fails. TLS in transit, SPF/DKIM/DMARC to stop spoofing, spam and virus scanning, attachment filtering.
Proxy Sits between clients and the internet: forwards requests on the client's behalf, caches responses, filters content and hides internal addresses. Response caching cuts upstream bandwidth as users grow; multiple proxies with load balancing scale out. Cache can serve content when the origin is unreachable — but the proxy itself is a single point of failure unless duplicated. Central content filtering and logging, plus hides internal IPs; but it can decrypt TLS, which raises privacy risk, and becomes a high-value target.
Web Serves HTTP/HTTPS content to clients; handles static files and passes dynamic requests to an application server; terminates TLS. Horizontal scaling behind a load balancer, CDN caching at the edge, virtual hosting of many sites per server. Health checks remove a failed server from the pool; redundant servers and automated redeployment. TLS, a WAF, patching, rate limiting and running as a low-privilege user.
Do not mix these up

DNS turns names into addresses. DHCP hands out addresses. A proxy forwards traffic on your behalf; a web server actually hosts the content. Examiners deliberately put DNS and DHCP in the same question because students swap them.

A2.2.3 Networking models: client–server vs peer-to-peer

Syllabus: compare and contrast the client–server and peer-to-peer models, including their benefits and drawbacks and real-world applications.

Client–server

Dedicated servers provide a service; clients request it. Roles are fixed and unequal.

Benefits

  • Centralised control: one place for accounts, permissions and policy.
  • Centralised backup, so data is not lost when one client dies.
  • Consistent, authoritative data — no conflicting copies.
  • Scales predictably (add servers, load-balance) and is easier to audit.

Drawbacks

  • Server hardware, licences and admin staff cost money.
  • The server is a single point of failure unless it is made redundant.
  • Bottleneck and dependency: no network, no service.

Peer-to-peer (P2P)

Every node is both client and server; peers share directly. Pure P2P has no coordinator; hybrid P2P uses a tracker or index server to find peers.

Benefits

  • Cheap — no dedicated server hardware or licences.
  • Scales in bandwidth: the more peers, the more sources to download from.
  • Resilient — no single point of failure; the network works while peers remain.

Drawbacks

  • Inconsistent availability — content disappears when peers go offline.
  • Hard to secure, control or audit; no central backup or permissions.
  • Variable performance; each peer's upload speed limits others.
  • Malware and unverified content spread easily.

Applications — and which model each uses

ApplicationModelReason
Web browsingClient–serverBrowsers request pages from web servers; content must be authoritative and centrally maintained.
EmailClient–serverMail servers relay and store messages (SMTP/IMAP); central storage keeps one copy and enables filtering.
Online bankingClient–serverTrust, regulation and audit demand one authoritative vault with strict permissions.
File sharingBothBitTorrent is P2P (many sources, scales with users); a school's shared drive is client–server (permissions, backup).
VoIPHybridA signalling server sets up the call, then media often flows peer-to-peer (as in WebRTC) to cut latency and server load.
BlockchainP2PMany nodes each hold and validate a copy of the ledger; no central authority to trust or to fail.
The classic 6-mark trap

"Compare and contrast" needs both similarities and differences. State the shared property (both let nodes exchange data; both need addressing and a physical medium), then alternate the differences point-for-point. Two benefits of each plus one limitation, each developed, is a safe full-mark structure.

A2.2.4 Network segmentation

Syllabus: explain the concepts and applications of network segmentation — for performance and security, to reduce congestion and to manage resources — including the uses and roles of segmenting, subnetting and VLANs.

Why segment at all

  • Performance: a broadcast stays inside its segment instead of being delivered to every host, which cuts congestion and wasted bandwidth.
  • Security: a compromise is contained to one segment; access between segments is then controlled by ACLs, a firewall or a router.
  • Resource management: apply QoS and bandwidth limits per group (e.g. video-conference traffic prioritised over guest Wi-Fi), and troubleshoot faults that are isolated to a segment.

The three mechanisms

SegmentingSubnettingVLAN
What it isThe general act of dividing a network into separate parts — physically (separate switches, cabling, media) or logically.Dividing one IP network into smaller subnets with a longer prefix/mask; each subnet gets its own network address, host range and broadcast address.A logical grouping of switch ports (or SSIDs) into separate broadcast domains on shared physical hardware, tagged with 802.1Q.
LayerAny — physical through application.Layer 3 (IP addressing).Layer 2 (switching).
How traffic crossesRouter or firewall between segments.Router or L3 switch routes between subnets.Trunk link to a router/L3 switch for inter-VLAN routing.
StrengthStrongest isolation — separate hardware cannot be reached by a Layer-2 attack.Better use of address space and smaller broadcast domains; clear policy boundary.No re-cabling: membership follows the configuration, so a device can move desk and keep its segment.
WeaknessExpensive in hardware and cabling; inflexible.Planning overhead; each subnet wastes addresses (network + broadcast).Misconfigured trunks allow VLAN hopping; still shares physical failure domains.

Worked subnetting example

192.168.10.0/24 → two subnets with a /25 mask (255.255.255.128). The borrowed bit splits the host range in half:

  • 192.168.10.0/25 — hosts .1–.126, broadcast .127
  • 192.168.10.128/25 — hosts .129–.254, broadcast .255

126 usable addresses per subnet instead of 254 in one flat network. The point of the question is never the arithmetic — it is that the router now becomes a policy enforcement point between the two halves.

Typical application answer

A school puts staff, students, guests and IoT devices (cameras, printers) on separate VLANs with separate subnets. Guests reach the internet only, so a compromised guest laptop cannot reach the student file server; IoT devices are blocked from initiating connections to client devices; staff get priority bandwidth. All of it runs on the same physical switches, so no additional cabling is needed, and a future change of policy is a configuration change rather than a rewiring job.

Technique Turning knowledge into marks

The content gets you halfway. These are the habits that convert it into marks.

DescribeState the features — what it is and what it does. No reasons needed. One mark per distinct, correct feature.
ExplainGive the mechanism and the consequence: "…because …, which means …". A cause with no effect rarely earns the second mark.
Compare / contrastSimilarities and differences, point against point. Use "whereas" so the contrast is explicit.
DiscussBoth sides plus a judgement tied to the scenario. Marks come from development, not from a longer list.

Three habits

  1. Name the scenario. "For a small office" / "in the hospital's ward" — quote the context in every factor you use. Unapplied factor lists are the single biggest mark loss on this topic.
  2. Count the marks, then count your points. Four marks is usually two points developed, or four distinct points. Do not write one long sentence and hope.
  3. Use the syllabus's own vocabulary. Reliability, transmission speed, scalability, data collisions, cost. Collision domain, broadcast domain, lease, trunk, ACL. Examiners are matching against the guide.
If you are stuck

Run the factor list. For any "should they use X?" question, march through reliability → speed → scalability → collisions → cost and you will always have five developed points ready.

Practice Ten questions with model answers

Answer closed book first. The model answers are written the way a mark scheme reads: one bullet, one mark.

1. A small design studio of 12 people wants a new wired LAN. Describe two factors that would lead you to recommend a star topology over a full mesh. 4
  • Cost (1): a star needs one switch and one cable run per desk, whereas a full mesh needs a cable and an interface between every pair of the 12 machines — 66 links (1) — which is far more hardware and installation cost for a studio that only needs shared file access.
  • Scalability (1): in a star a new designer is added with a single cable to a free switch port, while in a full mesh every new machine needs a new link to each existing machine, so the star keeps growth cheap as the studio hires (1).

Accept reliability or speed instead, provided the comparison is explicit and tied to the studio.

2. Explain why a full mesh is used in WAN backbones but rarely in a home network. 4
  • In a backbone, reliability is critical: a full mesh gives every node at least two independent paths, so traffic re-routes when a link fails and there is no single point of failure (1), which keeps many organisations' traffic flowing (1).
  • Mesh cost grows as n(n−1)/2: for a backbone with few, high-capacity sites the link count is manageable, but for a home with a handful of devices these links would be unnecessary cost and cabling (1).
  • A home network carries light traffic to one internet connection, so redundancy buys nothing — a single router (a star) is enough, and it is cheap and simple to install (1).
3. [HL] Describe the function of a proxy server and explain one security benefit to a school. 4
  • It sits between clients and the internet and forwards their requests, returning the response to the client (1).
  • It caches frequently requested pages, reducing bandwidth use and improving load times for repeated requests (1).
  • Security: all outbound traffic passes one point, so the school can filter and log which sites are reached and block categories, enforcing acceptable-use policy centrally (1).
  • It also hides internal IP addresses behind a single public address, so external hosts cannot address internal machines directly (1).
4. [HL] Explain how DHCP contributes to the scalability of a large network. 4
  • DHCP automatically assigns IP address, subnet mask, gateway and DNS to a host, so no administrator configures each device by hand (1).
  • Adding hundreds of devices — or connecting guests' phones — therefore takes no extra administrative effort, which is what allows the network to grow (1).
  • Scopes and reservations let one server serve several subnets, and leases return unused addresses to the pool so a limited address range is reused efficiently (1).
  • Running a second DHCP server for the same scopes means address assignment survives one server failing, so growth does not create a single point of failure (1).
5. A media company must share 20 TB of video between 60 editors. Compare client–server and peer-to-peer for this task. 6
  • Both models let the 60 editors exchange the same files over one network; the difference is where the files live and who controls them (1).
  • Client–server: one authoritative copy on a file server, so every editor works from current versions rather than conflicting copies (1).
  • Permissions and audit trails are centralised, which matters for confidential client footage, and backups are taken once (1).
  • The server is a single point of failure and a bandwidth bottleneck for 20 TB, so it needs RAID, redundancy and a fast link — a real cost (1).
  • P2P: cheap, and large files distribute well because each editor who has a chunk serves it to others, so bandwidth grows with the team (1).
  • But availability depends on editors' machines being online, there is no central backup or version control, and anyone with a copy can leak footage — so P2P is unsuitable here (1).
  • Judgement: client–server (with a fast server and redundancy) is the better fit; P2P only makes sense for distributing finished, non-confidential renders (1).
6. Explain how blockchain relies on the peer-to-peer model. 4
  • There is no central server: every participating node holds a copy of the ledger and acts as both client and server (1).
  • A new transaction is broadcast to peers, who each validate it against the ledger rules before accepting it (1).
  • Because many independent nodes hold the same record, altering history would require changing most copies, which provides tamper resistance without a trusted authority (1).
  • The network keeps working as peers join or leave, so there is no single point of failure — the reason the model is chosen over client–server (1).
7. Explain how subnetting a flat /24 office network into two /25 subnets reduces congestion. 4
  • A subnet defines a broadcast domain, so a broadcast from one host reaches only the hosts in its own /25 rather than all 254 (1).
  • Broadcast traffic is therefore halved in each segment, leaving more bandwidth for useful traffic (1).
  • Only traffic addressed to the other subnet crosses the router, so local traffic stays local and does not consume the link between segments (1).
  • Fewer hosts contending per segment also means less contention and fewer retransmissions, improving response times (1).
8. Explain the difference between a subnet and a VLAN. 4
  • A subnet is a Layer-3 division of an IP address range defined by a mask; membership follows the host's IP address (1).
  • A VLAN is a Layer-2 division of switch ports (or SSIDs); membership follows switch configuration, not the address the device happens to have (1).
  • Both create separate broadcast domains, so both reduce broadcast congestion and let policy be enforced between groups (1).
  • They are usually paired — one VLAN per subnet — because a device must be in the matching IP range to use its VLAN; a VLAN alone would still need routing to reach another VLAN (1).
9. Discuss whether a school should put guest Wi-Fi on a VLAN or on physically separate networking equipment. 6
  • VLAN: guests are isolated logically at no extra hardware cost, since the existing switches and access points carry a tagged guest network (1).
  • Policy is a configuration change — moving a room to the guest network, or adding an SSID, needs no re-cabling (1).
  • But guests share the same physical switches and uplinks, so a switch failure takes both networks down, and a misconfigured trunk risks VLAN hopping (1).
  • Physical separation: a compromise on the guest network cannot reach school hardware at all, and a fault or overload on one network cannot affect the other (1).
  • It costs a second set of switches, access points and cabling, plus separate management — significant for a school budget (1).
  • Judgement: a VLAN with ACLs blocking guest-to-internal traffic and a bandwidth cap gives adequate isolation for guest internet access at a fraction of the cost; physical separation is justified only if the guest network must carry sensitive traffic (1).
10. [HL] A company's website failed for two hours when its single DNS server went offline. Explain two changes that would improve reliability. 4
  • Run at least two authoritative name servers in different locations and list both in the domain's NS records, so resolvers use the other when one is unreachable (1), and the zone stays resolvable during maintenance or a site outage (1).
  • Use caching and sensible TTLs: resolvers keep answers for the TTL, so repeated queries are answered without contacting the server (1); a shorter TTL then lets a planned change fail over quickly, while caching absorbs short outages (1).
  • Also accept: anycast so queries reach the nearest healthy instance; monitoring and automatic failover.

Revision Traps and key terms

Five mistakes that cost marks

1. Saying mesh is "faster" — it is more reliable; more hops usually means more latency.  2. Swapping DNS and DHCP.  3. Treating VLANs and subnets as the same thing (Layer 2 vs Layer 3).  4. Giving generic factor lists without applying them to the scenario.  5. "Compare" answers with only differences.

Topology
The arrangement of links and nodes in a network.
Star topology
Every host connects to one central switch or access point.
Full / partial mesh
Every node connected to every other / to several others.
Hybrid topology
A combination of topologies, e.g. a meshed core with star branches.
Collision domain
The set of devices whose transmissions can collide; a switch port is its own domain.
Broadcast domain
The set of devices that receive a broadcast frame; bounded by a router or a VLAN.
DNS
Resolves domain names to IP addresses via a distributed hierarchy.
DHCP
Assigns IP configuration automatically for a lease period (DORA).
Proxy server
Forwards client requests, caches responses, filters and logs traffic.
Client–server
Dedicated servers provide services; clients request them.
Peer-to-peer
Nodes act as both client and server with no central authority.
Subnetting
Dividing an IP network into smaller subnets using a longer prefix/mask.
VLAN
A logical Layer-2 grouping of ports into its own broadcast domain, tagged with 802.1Q.
Trunk port
A link that carries traffic for multiple VLANs between switches.
ACL
Access control list — rules permitting or denying traffic between segments.

Progress Coverage checklist

Tick each one only when you can produce it from memory. Your ticks are saved in this browser.