Services → packet switching (virtual circuit vs. datagram) → IPv4 addressing → subnetting → DHCP, ARP → IPv4 header, fragmentation → NAT.

Practise subnetting until it’s automatic: it is the biggest calculation topic in the course.

Tip Subnet in four lines: block size = 256 - the interesting mask octet · network = the multiple of the block at or below the address · broadcast = next network - 1 · usable = network + 1 to broadcast - 1.

Seen again Ch 1 Part 2, slide 33: Network layer overview

Packetizing: wrap the payload in a packet at the source, unwrap at the destination, never change or use it (like the post office).

Routing = run routing protocols to BUILD the routing table (find best paths).

Forwarding = what the router DOES with each arriving packet: pick the output interface.

Memory trick Routing is planning the trip; forwarding is turning at each junction.

Seen again Ch 8, slide 3: Routing protocols fill the table

Two packet-switching approaches: virtual circuit (connection-oriented) and datagram (connectionless).

Virtual circuit: set up a connection first, then every packet follows the SAME path and carries a VC identifier (flow label).

Setup reserves resources in every switch. “Virtual” because the circuit is not dedicated wire: resources are shared.

Each switch maps (incoming port, label) → (outgoing port, new label).

Setup and teardown add delay, but no per-packet routing decisions.

Datagram (connectionless): each packet handled independently; packets may take different paths.

No setup, full addresses in every packet, may arrive out of order, robust (if a router dies, other packets reroute).

The router looks only at the destination address.

2³² ≈ 4.3 billion addresses.

Bit weights: 128 64 32 16 8 4 2 1.

Subtract the biggest weight that fits, write 1, continue.

Each octet is 0–255. Also watch for leading zeros (045) and more than four octets.

Prefix = network, suffix = host. Network address = host bits all 0.

A router has one IP address PER INTERFACE (one per network it touches).

Classful: classes A, B, C, D, E.

First byte: A 0–127 · B 128–191 · C 192–223 · D 224–239 · E 240–255.

Memory trick First-byte limits: 127 / 191 / 223 / 239 / 255.

Class A: /8, 126 usable networks, ~16.7 million hosts each. 127 = loopback.

Class B: /16, 16,384 networks, 65,534 hosts each.

Class C: /24, 254 hosts. Class D: multicast, no net/host split.

Private ranges: 10/8, 172.16–172.31, 192.168/16. Not routable on the Internet (NAT needed).

Seen again This deck, slide 84: Private addresses and NAT

Class A → keep 1 byte, zero the rest: 23.0.0.0.

Host bits all 0 → network address. All 1 → directed broadcast. Everything in between is usable.

Directed broadcast = net ID + host all 1s, sent to ANOTHER network. Routers forward it (often blocked: DoS risk).

Trap Directed broadcast (e.g. 192.168.10.255) CAN cross routers; limited broadcast 255.255.255.255 cannot.

Shortcut: mask 255 → copy the octet, mask 0 → write 0, otherwise AND in binary.

Tip Mask octets to know by heart: 128 · 192 · 224 · 240 · 248 · 252 · 254 · 255.

Sender ANDs the destination with its mask: same network → deliver directly; different → send to the default gateway.

Classful is dead: today addresses are classless (CIDR), written /n.

Fill whole octets of 255 (8 bits each), then the leftover bits go in the next octet.

Tip Mask octets to know by heart: 128 · 192 · 224 · 240 · 248 · 252 · 254 · 255.

Memorize this list.

1 AND 1 = 1, everything else 0.

Only the octet where the mask is not 255 or 0 needs binary work.

One organisation, one block, many departmental LANs → subnetting.

Benefits: less traffic (smaller broadcast domains), easier management, control over the address space, security.

Example: 192.168.1.0/24 into 5 subnets. 5 - 1 = 4 = 100 → 3 bits → /27 (255.255.255.224).

Weight: the last 1 of the mask sits at 2⁵ = 32 → subnets .0, .32, .64, .96, .128, .160, .192, .224.

Memory trick Bits needed = smallest n with 2ⁿ ≥ subnets. Block size = 256 - the mask octet.

The slide’s table is blank in the PDF: these are the answers. Each /27 has 32 addresses, 30 usable.

One /27 per floor; .160, .192 and .224 are kept for expansion.

Use the answers on slide 58.

141.14.0.0/16 split into 4 subnets with /18: .0.0, .64.0, .128.0, .192.0.

Four settings every host needs: IP address, subnet mask, default gateway, DNS server.

Manual configuration doesn’t work for diskless, newly booted or moved machines → DHCP.

Two networks joined by one router: 192.168.10.0/24 (A–D) and 192.168.20.0/24 (S).

Gateway for A–D: the router interface on their LAN (192.168.10.1). For S: 192.168.20.1.

IP: A → S on both sides (unchanged).

MAC from A: source A’s MAC, destination the router’s 192.168.10.1 interface. At S: source the router’s 192.168.20.1 interface, destination S’s MAC.

Seen again Ch 1 Part 2, slide 28: IP stays end to end, MAC changes per hop

/29 → block 8 → 35 is in 32–39 → network 211.56.77.32, broadcast 211.56.77.39.

Others: 172.16.192.0 and 220.1.2.160 (see slide 52).

Trap DHCP lives in the application layer even though it configures network-layer settings.

IP gets the packet to the right NEXT node logically, but the frame needs that node’s MAC address. ARP finds it.

Request = BROADCAST (“who has IP N2?”). Reply = UNICAST (“I do, my MAC is L2”).

Memory trick Ask everyone, answer one.

ARP is carried directly in an Ethernet frame (type 0x0806), not inside IP.

Seen again Ch 4, slide 8: The Ethernet type field

TTL: each router subtracts 1; 0 → drop (kills loops).

Protocol field: who gets the payload (TCP 6, UDP 17…). Checksum covers the HEADER only and is recomputed at every router.

Trap The IPv4 checksum does not cover the data.

ICMP 1, IGMP 2, TCP 6, UDP 17, OSPF 89.

Each network has its own frame format and size, so a router may have to cut a datagram to fit the next network.

Ethernet MTU = 1500 bytes (= Ethernet’s max data).

MTU = biggest payload a frame can carry. The datagram must be ≤ MTU, or it gets fragmented.

M = more fragments follow. D = don’t fragment (too big + D = 1 → dropped).

Same one’s-complement checksum as Chapter 3, over the header words only.

Seen again Ch 3 Part 1, slide 42: The Internet checksum

Always-on customers each need an address; there aren’t enough. NAT lets many private hosts share one public IP.

Inside: private 172.18.3.x. Outside: one public 200.24.5.8.

Out: the router rewrites the SOURCE (private → public). In: it rewrites the DESTINATION (public → private).

The table entry made on the way out tells the router where to send the reply.

NAT advantages: saves public addresses, change inside addresses or ISP freely, acts like a firewall (outsiders can’t start a connection in).