Segmentation
Segmentation turns one break-in into one room, not the building
Network segmentation limits lateral movement and reduces audit scope.
Picture the same break-in on two networks. On a flat network, one phished laptop can reach every machine: all twelve. That is how ransomware reaches every server in a night. Now split the network into zones: users, servers and payments, with firewalls between them that allow only what each zone needs. The same laptop reaches four machines. The servers and the payment systems are out of reach.
Ways to segment · Going deeper
VLANs, screened subnets, microsegmentation and air gaps
Segment with VLANs and subnets, a screened subnet (DMZ), microsegmentation, jump servers and air gaps.
Segment with VLANs and subnets, a screened subnet (DMZ), microsegmentation, jump servers and air gaps. Subnets and VLANs split one physical network into separate logical ones; traffic between them passes through a router or firewall where rules apply. Screened subnet (DMZ) a zone for internet-facing servers, between two firewalls, so a breached web server does not land inside the internal network. Microsegmentation rules around each workload, often enforced by software on the host or the hypervisor, so even two servers in the same zone cannot talk unless allowed. Jump server (bastion host) the one hardened, logged machine administrators use to reach a sensitive zone. Air gap no network connection at all, for the most critical systems such as some industrial controls. Removable media then becomes the path in. Cloud virtual networks, subnets and security groups do the same job; the principles are identical. Segmentation also limits compliance scope: only the zone holding card data is in scope for PCI DSS, if the boundary is proven.
- Subnets and VLANs split one physical network into separate logical ones; traffic between them passes through a router or firewall where rules apply.
- Screened subnet (DMZ) a zone for internet-facing servers, between two firewalls, so a breached web server does not land inside the internal network.
- Microsegmentation rules around each workload, often enforced by software on the host or the hypervisor, so even two servers in the same zone cannot talk unless allowed.
- Jump server (bastion host) the one hardened, logged machine administrators use to reach a sensitive zone.
- Air gap no network connection at all, for the most critical systems such as some industrial controls. Removable media then becomes the path in.
- Cloud virtual networks, subnets and security groups do the same job; the principles are identical.
Firewall rules
A firewall reads rules top to bottom; the first match decides
Firewall rules use first match and end in implicit deny. Stateful firewalls track connections.
A firewall decides with an ordered list. Here is a packet: a user’s laptop trying to reach a server over SSH, port twenty-two. Rule one allows web traffic to the web server: wrong destination. Rule two allows users to the app servers on eighty-four forty-three: wrong port. Rule three allows SSH, but only from the admin jump host: wrong source. Rule four denies everything. The packet is dropped. First match wins, and anything not allowed is denied.
Firewall types · Going deeper
From packet filters to next-generation and web application firewalls
Know stateless and stateful firewalls, proxies, NGFW, WAF, host-based firewalls and cloud security groups.
Know stateless and stateful firewalls, proxies, NGFW, WAF, host-based firewalls and cloud security groups. Packet filter (stateless) checks each packet’s addresses and ports alone. Fast, but replies need their own rules and it is easy to fool. Stateful firewall tracks connections, so a reply to an allowed request is let back in automatically. Proxy (application-level gateway) ends the connection and opens a new one, inspecting the application content in between. Next-generation firewall (NGFW) recognises applications and users, not just ports, decrypts TLS for inspection, and includes IPS features. Web application firewall (WAF) protects web apps from injection, XSS and similar at layer 7; can “virtually patch” a known flaw. Host-based firewall runs on each machine, so protection travels with a laptop. Cloud security groups and network ACLs security groups are stateful and attached to instances; network ACLs are stateless and attached to subnets.
- Packet filter (stateless) checks each packet’s addresses and ports alone. Fast, but replies need their own rules and it is easy to fool.
- Stateful firewall tracks connections, so a reply to an allowed request is let back in automatically.
- Proxy (application-level gateway) ends the connection and opens a new one, inspecting the application content in between.
- Next-generation firewall (NGFW) recognises applications and users, not just ports, decrypts TLS for inspection, and includes IPS features.
- Web application firewall (WAF) protects web apps from injection, XSS and similar at layer 7; can “virtually patch” a known flaw.
- Host-based firewall runs on each machine, so protection travels with a laptop.
- Cloud security groups and network ACLs security groups are stateful and attached to instances; network ACLs are stateless and attached to subnets.
IDS and IPS
An IDS tells you about an attack; an IPS can stop it
An IDS watches a copy of traffic and alerts; an IPS sits inline and blocks. Both use signature-based and anomaly-based detection.
Firewalls decide by address and port. Intrusion systems look inside the traffic. An intrusion detection system reads a copy of the traffic, from a mirror port or a tap. It raises an alert, but the packet has already arrived. It cannot break anything, and it cannot stop anything. An intrusion prevention system sits in the path. It can drop the malicious packet in real time. The price: a false alarm blocks real users, and you must decide whether it fails open or closed.
IDS and IPS · Going deeper
Most alerts are false alarms, and the misses are the danger
True positives, false positives, false negatives and true negatives; tuning trades one error for another.
No detector is perfect, and its mistakes come in two kinds. Take one week: ten thousand events, fifty of them real attacks. The system catches forty-five: true positives. And nine thousand six hundred and ninety-five benign events pass quietly. But five attacks slip by unseen: false negatives. The most dangerous cell, because nobody looks. And it raises two hundred and fifty-five false alarms. Only fifteen percent of alerts are real. Analysts drown. Tighten the rules and you get fewer false alarms, but more misses.
Zero trust
Zero trust checks every request; location earns nothing
Zero trust (NIST SP 800-207) uses a policy decision point and a policy enforcement point for every request.
The old model trusted anyone inside the network. Zero trust drops that idea. Priya, working from home, asks for the payroll app. Her request hits an enforcement point: a gateway or proxy. It asks a policy decision point. Who is she, did she pass MFA, is her laptop patched and encrypted, is this normal for her, how sensitive is the data? The answer: allow, for this session. She reaches the payroll app only, read only, and the check repeats as conditions change. Everything else stays out of reach. Being inside earns nothing.
Zero trust architecture · Going deeper
The parts of NIST SP 800-207, and how companies adopt it
NIST SP 800-207 defines the policy engine, policy administrator and policy enforcement point on a control plane and a data plane.
NIST SP 800-207 defines the policy engine, policy administrator and policy enforcement point on a control plane and a data plane. The parts. Policy engine (PE) makes the decision: grant, deny or revoke, using policy and signals. Policy administrator (PA) carries out the decision: sets up or tears down the session, issues tokens. Policy enforcement point (PEP) the gate in the traffic path that allows or blocks. Signals identity and MFA, device posture from EDR and device management, threat intelligence, activity logs, data classification. Control plane. Where decisions are made: the policy engine and administrator. Data plane. Where traffic flows: users, devices, enforcement points and resources. How it shows up in practice. Zero trust network access (ZTNA) replaces VPNs with per-application access. Secure access service edge (SASE) delivers ZTNA, secure web gateways and firewalls as one cloud service. Adoption is gradual: strong identity first, then device health, then microsegmentation.
The parts
- Policy engine (PE) makes the decision: grant, deny or revoke, using policy and signals.
- Policy administrator (PA) carries out the decision: sets up or tears down the session, issues tokens.
- Policy enforcement point (PEP) the gate in the traffic path that allows or blocks.
- Signals identity and MFA, device posture from EDR and device management, threat intelligence, activity logs, data classification.
Control plane
Where decisions are made: the policy engine and administrator.
Data plane
Where traffic flows: users, devices, enforcement points and resources.
How it shows up in practice
Zero trust network access (ZTNA) replaces VPNs with per-application access. Secure access service edge (SASE) delivers ZTNA, secure web gateways and firewalls as one cloud service. Adoption is gradual: strong identity first, then device health, then microsegmentation.
VPNs
A VPN protects the path across untrusted networks
A VPN uses IPsec or TLS. A full tunnel sends all traffic home; a split tunnel only company traffic.
On café Wi-Fi, anyone nearby might be listening. A VPN builds an encrypted tunnel across it. In a full tunnel, all traffic goes through the company: the apps, and your web browsing too, so company filtering applies everywhere. In a split tunnel, only company traffic uses the tunnel. Everything else goes direct: faster, and lighter on the company link, but outside the company’s protection. Either way, the VPN protects the path. It does not check who is on the other end, beyond the login.
DDoS
Against a flood, the defence is to be bigger than the flood
DDoS attacks are volumetric, protocol or application layer; scrubbing and anycast networks absorb them.
Some attacks do not break in at all. They simply overwhelm. A botnet of hijacked cameras and routers floods a target. The largest recorded, in 2025, reached thirty-one point four terabits a second. No single firewall can absorb that. So sites sit behind a scrubbing network spread across hundreds of locations. Each takes a share of the flood, and only clean traffic continues to the site. Floods come in three kinds: bandwidth, measured in bits; connection state, measured in packets; and costly requests, measured in requests per second.
Secure protocols
Every clear-text protocol has a secure twin, on its own port
Replace Telnet, FTP, HTTP, LDAP and SNMP v1/v2c with SSH, SFTP, HTTPS, LDAPS and SNMPv3.
Some protocols are older than the threats they face. Telnet, FTP and plain HTTP send everything, logins included, in clear text. Each has a secure replacement: SSH on twenty-two, SFTP or FTPS, HTTPS on four four three. Mail moves to five eight seven or four six five, and nine nine five and nine nine three. LDAP becomes LDAPS on six three six. SNMP version three adds authentication and encryption. And DNS can be encrypted on eight five three, or signed with DNSSEC. Clear text on a network is a finding.
Network attacks · Going deeper
ARP and DNS poisoning, rogue access points and Wi-Fi security
Know ARP poisoning, DNS poisoning, evil twin access points, VLAN hopping, WPA3 and 802.1X.
Know ARP poisoning, DNS poisoning, evil twin access points, VLAN hopping, WPA3 and 802.1X. Attacks on the local network. ARP poisoning the attacker answers “that IP address is at my hardware address”, putting themselves on the path. Defence: dynamic ARP inspection on switches. DNS poisoning false answers send users to the attacker’s server. Defences: DNSSEC, trusted resolvers, encrypted DNS. MAC flooding and VLAN hopping overwhelm a switch’s address table, or abuse trunk tagging to reach another VLAN. Defences: port security and disabling unused trunking. Rogue access point and evil twin an unauthorised access point plugged into the network, or a fake one copying a real network name to capture logins. Wireless security. Network access control (NAC) checks a device’s identity and health before letting it onto the network, often using 802.1X on wired ports too.
Attacks on the local network
- ARP poisoning the attacker answers “that IP address is at my hardware address”, putting themselves on the path. Defence: dynamic ARP inspection on switches.
- DNS poisoning false answers send users to the attacker’s server. Defences: DNSSEC, trusted resolvers, encrypted DNS.
- MAC flooding and VLAN hopping overwhelm a switch’s address table, or abuse trunk tagging to reach another VLAN. Defences: port security and disabling unused trunking.
- Rogue access point and evil twin an unauthorised access point plugged into the network, or a fake one copying a real network name to capture logins.
Wireless security
Lesson complete
You finished “Network security”
Nicely done.