1. IPv4 Addressing, Classless Inter-Domain Routing (CIDR) & VLSM

An IPv4 address consists of exactly 32 binary bits divided into four 8-bit octets separated by decimal periods. Under the original 1981 Internet Protocol specification (RFC 791), addresses were rigidly partitioned into classes: Class A (/8 prefix, 16.7M hosts), Class B (/16 prefix, 65,534 hosts), and Class C (/24 prefix, 254 hosts). This rigid allocation triggered severe address exhaustion and bloated core BGP routing tables.

In 1993, the IETF ratified Classless Inter-Domain Routing (CIDR) (RFC 1519, modernized in RFC 4632). CIDR decouples IP blocks from historical class boundaries, using an arbitrary prefix length \(p \in [0, 32]\):

The two reserved addresses within any standard subnet are:

• Network Identifier: All host bits set to 0. Used by routing tables to represent the entire subnet block.
• Directed Broadcast Identifier: All host bits set to 1. Used to send an IP datagram to all active nodes within that local broadcast domain.

Variable Length Subnet Masking (VLSM) enables network engineers to subdivide an assigned IP address space into subnets of differing sizes based on exact department or interface host counts, minimizing IPv4 address waste to near zero.

2. Packet Encapsulation, MTU Clamping & Path MTU Discovery (PMTUD)

Data moving down the OSI seven-layer protocol stack undergoes successive encapsulation: Application payload → Transport Layer segment (TCP/UDP) → Network Layer packet (IP) → Data Link Layer frame (Ethernet).

The standard Ethernet II frame has a Maximum Transmission Unit (MTU) of 1,500 bytes. Subtracting standard 20-byte IPv4 headers and 20-byte TCP headers yields the maximum payload carried per packet, known as the Maximum Segment Size (MSS):

When traffic traverses VPN tunnels (such as WireGuard, IPsec ESP, or GRE), encapsulation adds 32 to 73 bytes of outer headers. If the inner packet remains at 1,500 bytes, intermediate routers must either fragment the packet (degrading performance) or drop it if the Don't Fragment (DF=1) bit is set. Network devices prevent this via MSS Clamping, rewriting the SYN packet's MSS option to 1,360–1,400 bytes.

3. Bandwidth-Delay Product (BDP) & The Mathis Throughput Formula

In high-speed, high-latency networks (known as Long Fat Networks or LFNs), maximum TCP throughput is governed not merely by line rate, but by the volume of data that can be in transit before an acknowledgment (ACK) is required. This volume is the Bandwidth-Delay Product (BDP):

The original TCP header allocates only 16 bits for the Receive Window (RWIN), capping unscaled buffer size at 65,535 bytes (64 KB). Over a transcontinental link with an 80ms RTT, an unscaled window limits throughput:

even if the physical pipe is 10 Gbps! The TCP Window Scale option (RFC 7323) introduces a scale factor of up to 14 bits, expanding allowable window size up to 1 Gigabyte (2^30 bytes) to fully saturate multi-gigabit connections.

Under packet loss, TCP throughput degrades according to the classic Mathis Formula (1997):

where p represents packet loss probability and C ≈ 0.707. At just 1% packet loss on a 50ms link, TCP throughput collapses to less than 2 Mbps regardless of link bandwidth.

4. Recursive DNS Resolution & BGP Anycast Routing Architecture

Domain Name System (DNS) translation relies on a hierarchical distributed database: Stub Resolver → Recursive Resolver (e.g., 8.8.8.8, 1.1.1.1) → Root Nameservers → Top-Level Domain (TLD) Nameservers → Authoritative Nameservers.

Content Delivery Networks (CDNs) leverage BGP Anycast to announce identical IP address prefixes from hundreds of global Points of Presence (PoPs). Internet Service Provider (ISP) routers route traffic along the shortest AS-Path to the geographically closest edge server, reducing Round-Trip Time from 150–200ms to under 10–20ms and terminating TLS handshakes at the edge.

1. IPv4 Addressing, Classless Inter-Domain Routing (CIDR) & VLSM

An IPv4 address consists of exactly 32 binary bits divided into four 8-bit octets separated by decimal periods. Under the original 1981 Internet Protocol specification (RFC 791), addresses were rigidly partitioned into classes: Class A (/8 prefix, 16.7M hosts), Class B (/16 prefix, 65,534 hosts), and Class C (/24 prefix, 254 hosts). This rigid allocation triggered severe address exhaustion and bloated core BGP routing tables.

In 1993, the IETF ratified Classless Inter-Domain Routing (CIDR) (RFC 1519, modernized in RFC 4632). CIDR decouples IP blocks from historical class boundaries, using an arbitrary prefix length \(p \in [0, 32]\):

The two reserved addresses within any standard subnet are:

• Network Identifier: All host bits set to 0. Used by routing tables to represent the entire subnet block.
• Directed Broadcast Identifier: All host bits set to 1. Used to send an IP datagram to all active nodes within that local broadcast domain.

Variable Length Subnet Masking (VLSM) enables network engineers to subdivide an assigned IP address space into subnets of differing sizes based on exact department or interface host counts, minimizing IPv4 address waste to near zero.

2. Packet Encapsulation, MTU Clamping & Path MTU Discovery (PMTUD)

Data moving down the OSI seven-layer protocol stack undergoes successive encapsulation: Application payload → Transport Layer segment (TCP/UDP) → Network Layer packet (IP) → Data Link Layer frame (Ethernet).

The standard Ethernet II frame has a Maximum Transmission Unit (MTU) of 1,500 bytes. Subtracting standard 20-byte IPv4 headers and 20-byte TCP headers yields the maximum payload carried per packet, known as the Maximum Segment Size (MSS):

When traffic traverses VPN tunnels (such as WireGuard, IPsec ESP, or GRE), encapsulation adds 32 to 73 bytes of outer headers. If the inner packet remains at 1,500 bytes, intermediate routers must either fragment the packet (degrading performance) or drop it if the Don't Fragment (DF=1) bit is set. Network devices prevent this via MSS Clamping, rewriting the SYN packet's MSS option to 1,360–1,400 bytes.

3. Bandwidth-Delay Product (BDP) & The Mathis Throughput Formula

In high-speed, high-latency networks (known as Long Fat Networks or LFNs), maximum TCP throughput is governed not merely by line rate, but by the volume of data that can be in transit before an acknowledgment (ACK) is required. This volume is the Bandwidth-Delay Product (BDP):

The original TCP header allocates only 16 bits for the Receive Window (RWIN), capping unscaled buffer size at 65,535 bytes (64 KB). Over a transcontinental link with an 80ms RTT, an unscaled window limits throughput:

even if the physical pipe is 10 Gbps! The TCP Window Scale option (RFC 7323) introduces a scale factor of up to 14 bits, expanding allowable window size up to 1 Gigabyte (2^30 bytes) to fully saturate multi-gigabit connections.

Under packet loss, TCP throughput degrades according to the classic Mathis Formula (1997):

where p represents packet loss probability and C ≈ 0.707. At just 1% packet loss on a 50ms link, TCP throughput collapses to less than 2 Mbps regardless of link bandwidth.

4. Recursive DNS Resolution & BGP Anycast Routing Architecture

Domain Name System (DNS) translation relies on a hierarchical distributed database: Stub Resolver → Recursive Resolver (e.g., 8.8.8.8, 1.1.1.1) → Root Nameservers → Top-Level Domain (TLD) Nameservers → Authoritative Nameservers.

Content Delivery Networks (CDNs) leverage BGP Anycast to announce identical IP address prefixes from hundreds of global Points of Presence (PoPs). Internet Service Provider (ISP) routers route traffic along the shortest AS-Path to the geographically closest edge server, reducing Round-Trip Time from 150–200ms to under 10–20ms and terminating TLS handshakes at the edge.