IP Addressing (Layer 3)

Vendor-neutral Layer 3 background - IP addressing, classes, CIDR, subnet masks, and host math. This is the conceptual companion to the interactive Subnetting & CIDR explorer.

Layer 3 in one line

Layer 3 (the network layer) is where IP addressing and routing live: it gets a packet from one host to another across networks, using the IP address as the destination. Layers below it (2/1) move frames on a single link; Layer 3 stitches links into an internet.

An IPv4 address

A 32-bit number written as four dotted decimal octets (10.240.0.5), each 0-255. Every address splits into a network part (which network) and a host part (which machine on it). The split point is defined by the subnet mask.

Classful addressing (the old way)

Originally the split was fixed by class, determined by the leading bits / first octet:

Class	First octet	Default mask	Hosts per network	Purpose
A	0-127	/8 (255.0.0.0)	~16 million	Very large networks
B	128-191	/16 (255.255.0.0)	~65,000	Medium networks
C	192-223	/24 (255.255.255.0)	254	Small networks
D	224-239	-	-	Multicast
E	240-255	-	-	Experimental / reserved

Fixed classes waste addresses (a network needing 300 hosts had to take a whole Class B of 65,000). CIDR replaced this.

CIDR (the modern way)

CIDR = Classless Inter-Domain Routing. It drops fixed class boundaries and lets the network/host split fall at any bit, written as network/prefix:

10.0.0.0/24 - the /24 prefix means the first 24 bits are the network; the remaining 32 - 24 = 8 bits address hosts.

A smaller prefix number = more host bits = bigger range.

Subnet masks and host math

The subnet mask is just the prefix drawn as an address: the network bits are 1, the host bits are 0.

Prefix	Subnet mask	Host bits	Total addresses
/24	255.255.255.0	8	2⁸ = 256
/20	255.255.240.0	12	2¹² = 4,096
/16	255.255.0.0	16	2¹⁶ = 65,536
/8	255.0.0.0	24	2²⁴ = 16,777,216

Total addresses = 2^(32 − prefix)

Usable hosts = total − 2 - the network address (all host bits 0) and the broadcast address (all host bits 1) are not assignable.

So a /24 has 256 total and 254 usable. (GCP reserves 4 instead of 2 - see IP addresses.)

Worked example: sizing a /26 for a team

An office needs at least 50 usable host addresses. Check whether 10.0.1.0/26 works:

1. Prefix is /26, so host bits = 32 - 26 = 6.
2. Total addresses = 2^6 = 64.
3. Usable hosts = 64 - 2 = 62 (subtract the network address 10.0.1.0 and the broadcast address 10.0.1.63).
4. 62 ≥ 50, so /26 fits, with 12 addresses to spare for future growth.

If the team instead needed 65 hosts, /26 (62 usable) would be one address short - the next size up, /25, gives 2^7 - 2 = 126 usable hosts.

⚠

Off-by-one host counts around /31 and /32

The total - 2 formula assumes there is a dedicated network address and a dedicated broadcast address to subtract - true for any subnet with 2 or more host bits. It breaks down at the edges: a /32 has zero host bits, so it is a single address with no network/broadcast split at all (used to describe one specific host, e.g. a loopback or a route to one machine). A /31 has exactly one host bit, giving only two addresses total - and RFC 3021 defines both of them as usable host addresses on a point-to-point link (no network/broadcast reservation), so a /31 yields 2 usable hosts, not 2 - 2 = 0. Applying the generic formula to /31 or /32 without knowing this exception is a classic off-by-one mistake when sizing point-to-point WAN links or loopback routes.

Public vs private addresses

Private ranges (RFC 1918)

Not routable on the public internet; reused freely inside private networks and cloud VPCs:

• 10.0.0.0/8
• 172.16.0.0/12
• 192.168.0.0/16

Everything else is public (globally routable, must be allocated to you).

Prefix direction trips people up

/16 is bigger than /24. The number counts network bits - fewer network bits leaves more host bits, so a smaller number is a larger range.