IB COMPUTER SCIENCE | A2

NETWORKS REVISION

Connect network services, architecture, transmission and security through worked revision. Include the labelled HL material, then select flashcards to build your revision list.

Selections are retained, including cards hidden by filters. All selected cards remain in your revision list.

YOUR A2 REVISION ROUTE

Revise A2 in four connected passes: identify network services and infrastructure, compare architecture choices, trace transmission, then judge protection. Explain a process aloud, recreate a worked example and check recall with the flashcards.

The SL/HL switch hides the five HL-only objectives: A2.1.5, A2.2.2, A2.3.4, A2.4.2 and A2.4.3. Other material is common to both levels. Device mapping remains in the common A2.1.3 notes; the full TCP/IP model treatment is labelled HL.

Flashcards can be filtered by section and marked for further learning. Your selected revision list retains all chosen cards, including cards hidden by a level or section filter, while the page remains open.

A2.1.1 | NETWORK PURPOSE AND SCOPE

A network connects participating systems so they can communicate and share resources. Benefits include shared services, collaboration and managed access; costs include equipment, administration, dependencies and exposure to security risks. A connection does not itself guarantee permission or availability.

A LAN serves a local setting, a WAN links networks across a wider area, and a PAN connects devices in a personal setting. Ownership, technologies and scale vary, so do not define every LAN as one room or every WAN as the internet. A VPN is a logical connection over another network, often providing a protected tunnel; it is not simply a fourth geographical size.

ScenarioReasoned identification
Laptop and nearby wearablePAN: personal device connectivity
School building networksLAN sections, potentially linked by a wider campus design
Organisation connects distant officesWAN connectivity between sites
Staff uses a protected remote tunnelVPN over the underlying connection

A2.1.2 | DIGITAL INFRASTRUCTURES WORK TOGETHER

The internet is interconnected network infrastructure; the WWW is a service using web resources and protocols over networks. Cloud computing offers managed, on-demand resources with responsibilities depending on the service. A distributed system spreads work across cooperating components. Edge computing places suitable processing near the data source or action. Mobile networks support connectivity for moving devices.

A smart traffic system can analyse a sensor locally for a prompt response and send summaries to cloud services for wider coordination. This can reduce dependence on a remote round trip for every action, but edge devices still need power, maintenance and security. Distribution adds coordination and consistency challenges; it does not automatically improve every task.

ChoiceBenefit to connect to the scenarioLimit to remember
CloudShared scalable resources or managed servicesNetwork, cost, configuration and provider dependencies
EdgeNearby decisions and reduced upstream transferLimited local resources and dispersed management
Distributed processingWork sharing or resilience under suitable designCommunication, state and partial-failure complexity
Mobile connectivityAccess while movingCoverage, load, handover and radio conditions

A2.1.3 | MATCH DEVICE FUNCTION TO THE JOB

Device/functionKey roleCommon TCP/IP mapping
NICHost access to the local linkNetwork interface
Basic Ethernet switchLocal frame forwarding using MAC informationNetwork interface
Wireless access pointWireless link access and commonly LAN bridgingNetwork interface
ModemSignalling conversion for a suitable access mediumNetwork interface / physical access functions
Router / default gatewayIP forwarding between networksInternet
Protocol gatewayInterworking different protocols/representationsDepends on the translation, potentially application
Hardware firewallTraffic policy enforcementDepends on inspected fields, state and application capability

One broadband box can combine several roles. A basic switch does not choose the full internet route, and a router is not responsible for interpreting every web page. Map the operation rather than assuming every product belongs to exactly one layer.

A2.1.4 | TRANSPORT AND APPLICATION PROTOCOLS

ProtocolDescribe its serviceAvoid this confusion
TCPReliable ordered byte stream with recovery/control mechanismsDoes not itself encrypt data or preserve application write boundaries
UDPIndividual datagrams without built-in ordered reliable deliveryHigher layers can add recovery or security
HTTPWeb request/response meaning, methods and statusPOST is not automatically encrypted
HTTPSHTTP communication protected using TLSDoes not prove content/business honesty
DHCPHost configuration and address lease managementNot DNS name lookup

A common HTTP/1.1 or HTTP/2 arrangement uses TLS over TCP; HTTP/3 uses QUIC over UDP with additional transport/security mechanisms. The core comparison is each service’s role, not “UDP can never be reliable in any application”.

WORKED CHECK | DHCP IS NOT DNS

A laptop needs a new DHCPv4 allocation: Discover seeks a server, Offer proposes configuration, Request selects/requests it, and Acknowledgement confirms it under the protocol. An Offer alone is not a confirmed lease. Configuration can include address, subnet mask, gateway and a DNS resolver address.

DNS then answers name-related questions where needed. DHCP telling the laptop which resolver to use is different from DNS answering a query about portal.example. Valid existing leases, cached answers and established connections can be reused; not every page load repeats every setup exchange.

Explain the consequences: a DHCP outage may initially affect new allocations while a valid existing lease still works; an uncached name lookup can fail during a DNS outage even if the destination web service itself is healthy.

A2.1.5 | THE FOUR-LAYER TCP/IP MODEL HL ONLY

LayerResponsibilityExample
ApplicationService-specific messages and meaningHTTP, DNS, DHCP
TransportCommunication between software endpointsTCP/UDP; logical ports
InternetIP addressing and forwarding between networksIP packets and router decisions
Network interfaceCurrent link delivery and physical transmissionEthernet/Wi-Fi frames and signals

Sending uses lower-layer services and encapsulates carried data; receiving processes wrappers and passes data upward. TCP segment → IP packet → link frame is a useful teaching chain. A large page is not necessarily one packet.

IP is best effort. In a TCP case, endpoint sequence information, acknowledgements and retransmission support ordered reliable stream delivery. The four-layer model does not require TCP for every exchange or guarantee every transfer succeeds.

WORKED CHECK | ADDRESS, PORT AND LOCAL NEXT HOP

For a remote service, the host’s IP packet targets the service while its first Ethernet frame commonly targets the local router’s MAC address. A transport port helps identify the intended software endpoint. These are three different jobs.

At a normal routed boundary without NAT or tunnelling, endpoint IP addresses remain the same while link framing is replaced. Some IP fields change too, such as IPv4 TTL. Therefore “the packet is completely unchanged” and “the distant server’s MAC is used on every link” are both poor explanations.

A2.2.1 | STAR, MESH AND HYBRID

A switched star connects endpoints through a central switch. An endpoint cable fault normally isolates that endpoint, while a central switch failure interrupts communication through the star. A full mesh directly connects every pair in the considered set; a partial mesh provides selected interconnections. A hybrid combines arrangements, such as local stars with an interconnected core.

Alternative links help only when the remaining path is connected and forwarding mechanisms can use it. A hybrid’s redundant core does not repair an endpoint’s only failed cable. Redundancy must consider shared power, routes and equipment.

FactorUseful comparison
ReliabilityIdentify actual dependencies and usable alternatives
Transmission speedRates, shared links, devices and load matter; no universal winner
ScalabilityStar port/capacity limits versus steep full-mesh connection growth
CollisionsFull-duplex switched Ethernet avoids normal collisions; hubs/wireless differ
CostInclude installation, ports, maintenance and expertise

WORKED CHECK | MESH GROWTH AND A BOTTLENECK

A full mesh of six nodes needs 6 × 5 ÷ 2 = 15 two-way point-to-point links. Each node has five links. Adding a seventh node needs six additional links, giving 21. Division by two avoids counting A–B and B–A as separate physical connections in this model.

Twenty computers can each have 1 Gbit/s access links yet still compete for one 1 Gbit/s uplink to a shared service. The topology does not create unlimited capacity. Queueing or drops on full-duplex Ethernet can reflect congestion without normal Ethernet collisions.

For a campus, explain why manageable star access and selectively resilient core links may balance cost and availability better than a direct connection between every pair of laptops.

A2.2.2 | SIX SERVER ROLES HL ONLY

RoleFunctionImportant management concern
DNSAnswer name-related queriesCache limits, trusted records and service availability
DHCPManage configuration and leasesAdequate pools and coordinated failover
FileProvide network files/foldersPermissions, storage performance and recoverable backups
MailSubmit, transfer and store messagesQueues, mailbox capacity and safe account/relay policy
ProxyIntermediation for clients or backend servicesCorrect policy and private/public cache handling
WebHandle HTTP requests/responsesApplication/database dependencies and access controls

For each role compare function, scalability, reliability and security. A service can span hosts or share one host with other roles. Replication and redundant storage are not substitutes for protected historical backups.

A forward proxy acts for clients; a reverse proxy fronts services. SMTP supports submission/transfer; IMAP supports mailbox access. A certificate or permitted port does not prove all application content is safe.

A2.2.3 | CLIENT-SERVER VERSUS PEER-TO-PEER

Both models support network resource use. Client-server assigns provision to designated services. P2P participants can provide as well as consume resources. Models describe roles; star and mesh describe connection arrangements. A P2P application can operate over a physical star.

Client-server can simplify common policy, authoritative records and backups, but service dependencies need capacity and recovery planning. P2P can distribute provision and avoid a dedicated content host, but availability depends on reachable peers with the needed data, and coordination/trust remain necessary.

A web request or mailbox access is commonly client-server. File sharing can use either model. VoIP can combine central signalling and direct peer media; blockchain nodes may exchange P2P while a wallet uses a central API. Classify the specified interaction.

FactorClient-serverP2P
ControlCommon designated service policyPotentially distributed policy/coordination
GrowthProvision service and dependent capacityNew peers can add capacity and demand
FailureSingle service without failover can interrupt accessLast usable copy going offline can remove a resource
SecurityCentral controls help but central compromise mattersDistributed trust and endpoint controls still needed

A2.2.4 | SEGMENTATION, SUBNETS AND VLANS

Segmentation is the broader design of smaller sections and boundaries. A VLAN creates a logical Layer 2 LAN/broadcast domain on switching infrastructure. Subnetting divides IP address space using prefixes. A common design pairs one subnet with one VLAN, but their functions are distinct.

Access ports commonly attach endpoints to one data VLAN; trunks can carry multiple VLANs while preserving membership. A router, Layer 3 switch or suitable firewall supplies inter-VLAN routing. Routing makes a path possible; ACL/firewall policy controls which paths are permitted.

A student broadcast is not normally bridged into the staff VLAN, reducing irrelevant local traffic. Shared uplinks and internet capacity remain shared. A VLAN is not encryption and does not automatically prevent guest-to-guest communication without additional controls.

WORKED CHECK | SPLIT A /24 AND CONTROL A PATH

BlockOrdinary IPv4 interface rangeDirected broadcast
192.168.10.0/25192.168.10.1–126192.168.10.127
192.168.10.128/25192.168.10.129–254192.168.10.255

Each example has 128 total addresses, with 126 ordinary interface positions after excluding network/broadcast; a gateway consumes one such position. .126 and .129 are in different /25 subnets even though the first three decimal numbers match.

If student and service subnets occupy separate VLANs, a permitted HTTPS route can connect them while guest-to-service access is denied. The original student Layer 2 broadcast does not become a broadcast in every destination VLAN. A DHCP relay may be needed to reach a DHCP service outside the client’s broadcast domain.

A2.3.1 | VERSION, SCOPE AND ASSIGNMENT

IPv4 addresses have 32 bits and commonly use four decimal octets, each 0–255. IPv6 has 128 bits and full notation uses eight hexadecimal groups. Leading zeros inside IPv6 groups can be omitted, and one zero run can use ::. 2001:db8::24 expands to 2001:0db8:0000:0000:0000:0000:0000:0024.

Public/private scope and static/dynamic assignment are independent. A private printer address may be static while a private laptop address is leased dynamically. A provider can allocate a public address statically or dynamically. Repeatedly receiving the same address does not prove static configuration.

Private IPv4 ranges are 10.0.0.0/8, 172.16.0.0/12 and 192.168.0.0/16. Not all 172 addresses are private, and not every non-private address is ordinary public unicast: loopback and documentation ranges are examples of special use.

WORKED CHECK | NAPT AND IPV6 SCOPE

Internal TCP endpointIllustrative translated external endpoint
192.168.10.24:51000203.0.113.8:62001
192.168.10.25:51000203.0.113.8:62002

NAPT translates addresses and ports, letting internal connections share external IPv4 addressing. The separate external ports select the illustrated return mappings. This conserves public IPv4 use; it does not create more 32-bit patterns or encrypt content. 203.0.113.8 is a reserved documentation example, not a live public address.

IPv6 includes global unicast, link-local fe80::/10 and unique-local fc00::/7 categories. It can use automatic configuration such as SLAAC as well as other supported arrangements. Its large space reduces address-conservation NAT need, but neither removes firewall policy nor guarantees every application is faster.

A2.3.2 | COMPARE TRANSMISSION MEDIA

FactorTwisted pairFibre opticWireless
SignalElectrical over paired copperLight through an optical pathRadio in the Wi-Fi examples
Capacity/rangeDepends on category/standard; many office channels up to 100 mSuitable optics/fibre support high-capacity long linksDepends strongly on technology, environment and shared airtime
Installation/costRoutes, terminations and tests; often economical locallyCompatible optics, clean connectors and sometimes specialist handlingCoverage/channel planning, AP power and backhaul
Interference/attenuationElectrical noise and losses; twists/shielding helpOptical path immune to EMI, but optical loss existsPath loss, obstructions and competing transmissions
Reliability/securityPredictable within limits; protect cable/ports and dataPhysical/optical faults still matter; not untappable or encrypted by defaultMobility benefit; needs sufficient coverage and strong wireless protection

Bandwidth/capacity is not achieved throughput; latency is delay; attenuation is signal weakening. A campus may choose copper for nearby fixed PCs, fibre for a 250 m inter-building backbone and wireless for moving tablets. Explain why each choice fits instead of declaring one medium universally best.

A2.3.3 | PACKET SWITCHING AS A CONNECTED PROCESS

Application data is carried in suitable units, with control information supporting delivery. In a TCP example, segments are carried by IP packets and local link frames. IP addressing supports forwarding, while TCP has separate byte sequence, endpoint and recovery information.

Local switches forward frames within the relevant VLAN. Routers use destination IP and installed forwarding information to select successive next hops. Different transfers share links, so packets may queue or be lost. Paths can differ, but packets of one flow often follow the same stable path.

At the destination, TCP can reconstruct an ordered byte stream and recover missing ranges where possible. IP itself is best effort, and UDP applications do not inherit TCP recovery automatically. IP fragment reassembly is a separate mechanism from TCP stream ordering.

WORKED CHECK | OUT-OF-ORDER BYTES

Observed arrivalTCP contiguous bytes availableLater bytes waiting
NET at relative bytes 0–2NETNone
KS! at relative bytes 6–8NETKS!
WOR at relative bytes 3–5NETWORKS!None

The receiver uses byte-position information rather than blindly joining arrival order. If WOR is lost, recovery is needed before the gap is filled; a later retransmission may provide it. The labels are teaching offsets, not a universal packet-number field in IP.

Acknowledgements report transport progress, not proof that a document was saved or a payment completed. An unrecoverable failure can still break a TCP connection.

A2.3.4 | STATIC AND DYNAMIC ROUTING HL ONLY

Static routes are explicit administrator-defined instructions for destination prefixes and next hops. Dynamic routing uses configured protocol exchanges and selection rules to learn/update routes. Both feed installed forwarding; neither asks a human or reruns a complete route exchange for each application packet.

Static can suit a small stable single-exit section with little protocol overhead, but remote changes can require manual maintenance. Dynamic learning often helps a changing multipath campus, but adds configuration complexity, update traffic, state and calculation work.

Convergence includes detection, information changes, route selection and forwarding updates. It takes time and can lose in-flight traffic. Dynamic does not always mean fastest, and basic static routes do not automatically discover every alternate path. Configured tracking/backup can change static behaviour.

WORKED CHECK | ROUTE MATCH AND FAILURE HL ONLY

Installed destinationInstruction
10.20.0.0/16Via R2
10.20.30.0/24Via R3
0.0.0.0/0Via R5

10.20.30.8 uses the more-specific /24 instruction, while 10.20.99.8 uses /16. Prefix specificity is distinct from comparing metrics among candidate routes.

If a remote R2–R4 link fails while R1 can still reach R2, a static instruction without tracking/backup may keep sending toward a broken path until changed. Dynamic learning can select an available permitted alternative after convergence. Correct return routes and policy are still required.

Compare configuration, maintenance, complexity, resource usage, convergence, scalability and network size; static and dynamic can coexist.

A2.4.1 | FIREWALL EFFECTIVENESS

Firewalls enforce traffic policy using available fields, connection state and supported inspection features. Allowlists/whitelists identify approved paths; blocklists/blacklists identify denied items. A first-match ruleset needs careful ordering because a broad early permit can shadow a later deny.

Stateful inspection can validate return traffic for an allowed connection without opening arbitrary inbound access. An ACK flag alone does not prove a trusted reply. Outbound filtering also matters, but allowed paths may still carry harmful data.

Strengths include reduced reachability, consistent boundary policy and useful logs. Limits include bypassed internal paths, allowed/hidden threats, false decisions, misconfiguration and finite capacity. Large upstream saturation may need upstream mitigation. NAT can reduce straightforward unsolicited reachability in common IPv4 sharing but is not encryption or a complete firewall policy.

WORKED CHECK | DISCUSS A FIREWALL PROPOSAL

A school permits HTTPS browsing but denies guest-to-staff connections. Explain the benefit: required learning works while unnecessary targets are less reachable. Then state the limit: a permitted encrypted site can still deceive a user or serve harmful content, and local traffic bypassing the perimeter is not inspected there.

A supported judgement is that a properly placed and maintained firewall is an important access-control layer, combined with patching, MFA, endpoint security, safe applications, training and recovery. A valid policy depends on reviewed exceptions, useful monitoring and capacity, not simply owning a firewall.

Use “discuss” to develop strengths and limitations in context and reach a reasoned conclusion, rather than only defining packet filtering.

A2.4.2 | VULNERABILITIES AND ATTACK MECHANISMS HL ONLY

ConceptDefining feature
DDoSDistributed demand exhausts service or supporting resources
Insecure protocolsCommunication lacks necessary confidentiality/integrity/authentication
MalwareSoftware performs harmful or unauthorised activity
MitMUnauthorised intermediary can interfere where protections fail
PhishingDeception induces disclosure or unsafe action
SQL injectionInput alters database instructions through a vulnerable application
XSSUntrusted content executes in browser site context
Unpatched softwareApplicable available fixes have not been installed
Weak authenticationInadequate identity checks or credentials
Zero-day exploitUses a previously unknown/unfixed flaw with little preparation opportunity

A vulnerability is a weakness; an exploit uses one; an attack is an attempt at harm. A symptom such as slowness is not conclusive proof. Describe mechanism, processing location and a plausible confidentiality, integrity or availability impact.

WORKED CHECK | INJECTION, IDENTITY AND PATCH STATUS HL ONLY

SQL injection concerns unsafe database instruction construction. XSS concerns browser execution in a trusted site context. Both mishandle untrusted data but require context-specific defences. Not every textbox containing SQL words proves injection, and not every XSS can read every cookie.

Phishing can lead to malware and credential theft, but the terms identify different stages. MitM concerns an unauthorised path intermediary, not simply a convincing message. Correctly authenticated TLS resists relevant interception; it does not make a phishing site honest.

A known flaw with an available unapplied fix differs from a previously unknown/unfixed zero-day situation. Authentication checks identity; authorisation controls actions. A legitimate user reading forbidden records may indicate faulty authorisation rather than weak login.

A2.4.3 | COUNTERMEASURES BY PURPOSE HL ONLY

ControlFunction and limit
CSPBrowser resource/execution policy; complements safe rendering
Password policy / MFALong unique credentials and different factor types; secure recovery matters
DDoS mitigationFilter/absorb demand at useful locations; classification/capacity limits remain
Email filteringReject/quarantine/warn on suspicious evidence; can miss deception
Encrypted protocols / certificatesProtect and authenticate connection endpoints; not business honesty
Input validationCheck allowed data requirements; also use parameterised SQL and safe output handling
IDS / IPSDetect and alert / preventive action under policy; false decisions possible
UpdatesDeploy/verify fixes; not every future flaw disappears
VPNProtect a defined tunnel; onward paths and endpoints still need controls
Wireless securityStrong supported authentication/encryption; MAC lists are limited
Testing / trainingAuthorised checks, remediation/retesting and safe verification/reporting

SSL is an obsolete protocol family; “SSL certificate” is still a common label for certificates used with modern TLS. MAC filtering means Media Access Control addresses, which can be imitated or randomised; an address list is not strong identity assurance.

WORKED CHECK | DETECTION IS NOT AUTOMATIC PREVENTION HL ONLY

A passive IDS flags suspicious traffic and alerts without automatically blocking the original exchange. An inline IPS can drop flagged traffic under its policy. If a benign exchange is falsely flagged, the IPS can interrupt legitimate work; if harmful traffic is missed, neither model’s detector triggers the intended response.

MFA needs different factor types, not two passwords. Parameterised queries separate SQL structure and values, whereas validation checks data requirements and output encoding prevents unintended browser interpretation. A VPN protects between its endpoints, not every possible path.

Modern password guidance emphasises length, uniqueness and rejecting compromised/common choices rather than arbitrary composition or routine changes without a reason. Training and testing must feed real improvements rather than a one-time checklist.

A2.4.4 | SYMMETRIC, ASYMMETRIC AND SIGNATURE ROLES

GoalAppropriate key relationship
Symmetric protectionShared secret key relationships; traffic may use derived directional keys
Encrypt for Bob in a suitable public-key schemeBob’s public encryption key encrypts; Bob’s private key decrypts
Alice signs specified dataAlice’s private signing key signs; her public verification key verifies
Authenticate a public-key bindingCertificate and relevant trust/name/signature checks
Modern authenticated key establishmentKey agreement plus authentication derives traffic secrets

Algorithms specify procedures; keys control operations. Encoding such as Base64 is not secrecy, and hashing is not reversible encryption. A signature does not hide the message and is not universally “private-key encryption”. Not all public-key algorithms support the same roles.

WORKED CHECK | CERTIFICATE TO SECURE TRAFFIC

A certificate publicly binds identity information and a public key, with issuer/validity/other fields protected by a signature. The client checks the requested name, acceptable trust path, time and relevant constraints/status requirements. The server must prove control of the associated private key; copying its public certificate is insufficient.

In a fresh modern certificate-authenticated ephemeral TLS key-agreement example, exchange information and authentication support deriving traffic keys. Symmetric authenticated encryption then protects application traffic. Do not claim every modern TLS connection sends a session key encrypted with the certificate key.

Validated TLS protects its connection endpoints, not business honesty, every user action or every onward connection. Private keys are not inside public certificates, and authentication does not replace authorisation.

A2.4.4 | KEY LIFECYCLE AND PROTECTION BOUNDARIES

Key management covers secure generation, authenticated establishment, controlled storage/use, replacement, recovery where justified and retirement. Publicly leaking a private or shared secret can defeat a sound algorithm. Losing the only storage-decryption key can make records unrecoverable.

Compromise response can include replacing keys/certificates, revocation under the relevant system and investigation. Renewal and key rotation are not identical. Suitable ephemeral protocols can provide forward secrecy, but this does not undo stolen plaintext or endpoint compromise.

Encryption in transit protects an exchange; at rest protects stored representations. Both may use similar algorithm families but have different scope and key-management needs. Database encryption does not automatically prevent misuse through an authorised query account.

COMMON MISTAKES | CONNECT THE DISTINCTIONS

ConfusionUseful correction
Internet / WWWInfrastructure / web service
DHCP / DNSHost configuration / name information
Switch / routerLocal frame forwarding / IP network forwarding
Topology / modelConnections / service-provider roles
Subnet / VLANIP prefix / logical Layer 2 LAN
Dynamic IP / dynamic routingAddress assignment / route learning
Bandwidth / throughputCapacity / useful achieved rate
IP delivery / TCP recoveryBest effort / transport ordering and recovery
NAT / firewallTranslation / traffic permission policy
Public certificate / private keyDistributable signed binding / protected secret
Encryption / authorisationProtection of representations/exchanges / permitted actions

EXAM COMMUNICATION | USE THE COMMAND WORD

Describe: give main features and roles, using precise examples. Explain: state how a mechanism causes an outcome using because, therefore or resulting in. Compare: use the same factor for alternatives and identify relevant similarities/differences. Discuss: develop strengths and limits in context, then reach a supported judgement.

For topology, media and routing questions, explicitly cover the factors named in the objective. For a process, identify where each step occurs rather than attributing every action to “the network”. Recreate worked examples before checking notes, then select flashcards for the ideas you cannot explain unaided.