5 min read
On this page

Internet Architecture

Design Principles

The Internet's architecture emerged from a set of guiding principles rather than a single monolithic design. These principles shaped a network that is decentralized, resilient, and extensible.

Key Architectural Principles

| Principle | Description | |-----------|-------------| | Packet switching | Data divided into independent packets routed individually | | Layered protocols | Functionality separated into layers (TCP/IP model) | | Decentralization | No single point of control or failure | | Heterogeneity support | Interconnect diverse networks transparently | | Statistical multiplexing | Share bandwidth dynamically across flows | | Open standards | Publicly available protocol specifications |

The End-to-End Argument

Formulated by Saltzer, Reed, and Clark (1984), the end-to-end argument is the most influential design principle in Internet architecture.

Core claim: Functions that require knowledge of application semantics should be placed at the endpoints, not within the network core.

Application logic at endpoints:
  [Host A] ---- simple routers ---- [Host B]
     |          (forward only)          |
     |                                  |
  reliability                       reliability
  encryption                        encryption
  error recovery                    error recovery

Rationale

  • The network cannot implement end-to-end guarantees correctly without endpoint participation (e.g., reliable delivery still needs end-to-end acknowledgment even if each hop is reliable).
  • Placing logic in the core duplicates effort and penalizes applications that do not need it.
  • Keeping the core simple maximizes generality.

Violations and Tensions

  • NAT breaks end-to-end connectivity and complicates peer-to-peer protocols.
  • Firewalls and middleboxes inspect and modify traffic mid-path.
  • CDNs cache content inside the network, improving performance but adding complexity.
  • QoS mechanisms add per-flow state within the network.

Fate Sharing

Fate sharing dictates that state associated with a communication should be stored at the endpoints involved, so that state is only lost when an endpoint fails (which would end the communication anyway).

  • TCP connection state lives at sender and receiver, not in routers.
  • If a router reboots, ongoing connections survive (soft state in routers rebuilds via routing protocols).
  • Contrast with circuit-switched networks where switch failure destroys in-progress calls.

Internet Governance

IETF (Internet Engineering Task Force)

  • Develops Internet standards through open, consensus-driven process.
  • Documents published as RFCs (Request for Comments).
  • Organized into working groups under areas (Routing, Security, Transport, etc.).
  • "Rough consensus and running code" as decision-making philosophy.

ICANN (Internet Corporation for Assigned Names and Numbers)

  • Coordinates DNS root zone, IP address allocation, and protocol parameter assignment.
  • Delegates IP allocation to Regional Internet Registries (RIRs): ARIN, RIPE NCC, APNIC, LACNIC, AFRINIC.
  • Manages gTLD and ccTLD policies.

Other Bodies

| Organization | Role | |-------------|------| | IANA | Protocol parameter registry (operated by ICANN) | | IEEE | Link-layer standards (802.x) | | ITU-T | Telecommunications standards | | W3C | Web standards (HTTP semantics, HTML) | | ISOC | Internet Society, organizational home for IETF |

Autonomous Systems (AS)

An Autonomous System is a network or group of networks under a single administrative domain with a unified routing policy, identified by a 32-bit AS Number (ASN).

AS Types

| Type | Description | Example | |------|-------------|---------| | Stub AS | Single connection to one provider | Small enterprise | | Multi-homed AS | Connected to multiple providers, does not transit | Large enterprise | | Transit AS | Carries traffic between other ASes | Tier-1/2 ISPs |

AS Interconnection

Tier 1 ISPs (settlement-free peering among themselves)
   |              |              |
Tier 2 ISPs    Tier 2 ISPs    Tier 2 ISPs
   |              |
Tier 3 / Stub ASes

The traditional hierarchy is flattening as large content providers (Google, Netflix) peer directly with access ISPs.

Internet Exchange Points (IXPs)

IXPs are physical facilities where multiple ASes interconnect and exchange traffic, reducing latency and transit costs.

IXP Architecture

       IXP Switching Fabric (Layer 2)
      /      |       |       \
   AS-A    AS-B    AS-C    AS-D
   (ISP)  (CDN)  (ISP)  (Enterprise)
  • Participants connect to a shared switching fabric (typically Ethernet).
  • Each participant has a port on the IXP switch and a router.
  • BGP sessions established bilaterally or via a route server.

Route Servers

  • Simplify peering by acting as a BGP reflector at the IXP.
  • A participant peers with the route server instead of establishing N-1 bilateral sessions.
  • Route servers apply import/export policies on behalf of participants.

Major IXPs

DE-CIX (Frankfurt), AMS-IX (Amsterdam), LINX (London), Equinix IX (global). Some handle multiple Tbps of peak traffic.

Peering vs. Transit

Peering

  • Two ASes agree to exchange traffic destined for each other's networks (and customers) at no cost (settlement-free) or with partial payment (paid peering).
  • Neither AS provides access to the rest of the Internet through peering.
  • Requires roughly balanced traffic ratios in settlement-free agreements.

Transit

  • A customer AS pays a provider AS for full Internet reachability.
  • The provider advertises the customer's prefixes to the rest of the Internet.
  • The provider forwards traffic to/from the customer to any destination.

Peering Decision Factors

  • Traffic volume and ratio between networks.
  • Geographic overlap of networks.
  • Cost of transit vs. cost of peering infrastructure.
  • Strategic considerations (keeping traffic local, reducing latency).

Content Delivery Architecture

Traditional CDN Model

Origin Server
     |
CDN Control Plane (DNS-based or anycast request routing)
     |
  +--+--+--+--+
  |  |  |  |  |
 Edge servers distributed globally
  |  |  |  |  |
  End users (served from nearest edge)

Request Routing Mechanisms

| Method | How It Works | |--------|-------------| | DNS-based | CDN's authoritative DNS returns IP of nearest edge server | | Anycast | Same IP announced from multiple locations; BGP routes to nearest | | HTTP redirect | Origin returns 302 to edge URL | | URL rewriting | Application embeds edge-specific URLs |

Modern Content Delivery Trends

  • Embedded caches: ISPs host CDN nodes inside their network (Netflix Open Connect, Google GGC).
  • Edge computing: CDN nodes run application logic (Cloudflare Workers, Lambda@Edge).
  • QUIC/HTTP3: Reduce connection setup latency for content delivery.
  • Multi-CDN: Content providers use multiple CDNs simultaneously for resilience and performance.

Impact on Internet Architecture

Content providers now generate the majority of Internet traffic and increasingly bypass traditional transit relationships:

Before:  User → Access ISP → Transit ISP → Content Provider
Now:     User → Access ISP → [IXP or private peering] → Content Provider

This flattening reduces the average AS-path length and concentrates traffic on fewer, larger networks, fundamentally reshaping the economic and topological structure of the Internet.