Internet Backbone Architecture

When you stream a video from a server in Ireland while sitting in Los Angeles, your data travels through a complex physical and logical infrastructure that spans continents and oceans. This infrastructure — the internet backbone — consists of ultra-high-capacity fiber-optic cables, powerful routing equipment, and carefully negotiated business agreements. Understanding how it is structured reveals why some routes are fast, why the internet is resilient to failures, and why internet economics work the way they do.

The Tiered Network Model

The internet is organized into a hierarchy of networks called autonomous systems (ASes). These are grouped into three tiers based on their connectivity and business relationships.

Tier 1: The Core

Tier 1 networks are the backbone carriers. They have a unique characteristic: they can reach every other network on the internet without paying anyone for transit. They achieve this through settlement-free peering — mutual agreements with other Tier 1 networks to exchange traffic without money changing hands.

As of 2024, there are approximately 15-20 Tier 1 networks globally:

Network	Notable Reach
AT&T (AS7018)	North America, global
Verizon/AS701	North America, transatlantic
CenturyLink/Lumen (AS3356)	Global
NTT Communications (AS2914)	Global, strong Asia presence
Telia Carrier (AS1299)	Europe, global
GTT Communications (AS3257)	Global
Cogent (AS174)	North America, Europe
Hurricane Electric (AS6939)	Global, extensive peering
Deutsche Telekom (AS3320)	Europe
Telecom Italia Sparkle (AS6762)	Europe, Mediterranean

These networks own or lease the physical fiber-optic cables that cross continents and oceans. They operate Points of Presence (PoPs) in major cities — large co-location facilities housing routers, switches, and cross-connect infrastructure.

Tier 2: Regional Carriers

Tier 2 networks occupy the middle of the hierarchy. They peer for free with some networks but pay for transit to reach the full internet. Most national ISPs, regional carriers, and large enterprise networks fall into this category.

A European ISP, for example, might: - Peer for free with local content networks (Netflix, Google) at an IXP - Peer for free with neighboring ISPs of similar size - Pay a Tier 1 carrier for upstream transit to reach networks it cannot peer with directly

Tier 3: Access Networks

Tier 3 networks are the last-mile providers: residential ISPs, mobile carriers, and local internet service providers. They typically pay Tier 2 or Tier 1 networks for all of their upstream transit, and the cost of that transit is embedded in the price of your monthly internet subscription.

Optical Fiber: The Physical Layer

The backbone runs almost entirely on single-mode optical fiber — hair-thin strands of pure silica glass that carry light signals. Light travels through fiber with extremely low attenuation (signal loss), making it ideal for long-distance communication.

How Fiber Carries Data

A laser (or LED) at the transmitter pulses on and off at billions of times per second. These pulses encode binary data as presence or absence of light. At the receiver, a photodetector converts light back to electrical signals and reconstructs the data.

Modern single-mode fiber has an attenuation rate of approximately 0.2 dB/km. This means a signal loses about 5% of its power every kilometer. Over long distances, optical amplifiers (EDFA — Erbium-Doped Fiber Amplifiers) are placed every 80-100 km to boost the signal without converting it back to electrical form.

DWDM: Dense Wavelength Division Multiplexing

A single fiber strand can carry multiple data streams simultaneously by using different wavelengths (colors) of light. This technique — Dense Wavelength Division Multiplexing (DWDM) — dramatically multiplies the capacity of a physical fiber.

A modern DWDM system might carry: - 80 wavelength channels on a single fiber pair - Each channel running at 100 Gbps or 400 Gbps - Total capacity: 8 Tbps per fiber pair (and improving)

Because fiber cables contain many fiber pairs — submarine cables often contain 8-16 fiber pairs, and terrestrial cables can contain 864 or more fibers — the aggregate capacity of a single cable route can reach hundreds of terabits per second.

Dark Fiber

Not all fiber in the ground carries traffic. Dark fiber refers to installed fiber strands that are not yet lit (no optical equipment attached). When a carrier lays a new cable, it typically installs far more fiber than it currently needs — the cost of trenching and laying conduit dominates, so adding extra fiber pairs is cheap.

Dark fiber can be: - Leased to other carriers or enterprises who want to light it with their own equipment - Held in reserve for future capacity growth - Used as a strategic asset — organizations with dark fiber can increase capacity without new construction

Many municipalities, universities, and enterprises lease dark fiber from carriers to build private high-speed networks without paying per-megabit transit fees.

Backbone Capacity Growth

The history of backbone capacity is a story of exponential growth driven by advances in DWDM technology.

Year	Typical Backbone Link Speed
1988	1.5 Mbps (T1)
1991	45 Mbps (T3/DS3)
1995	155 Mbps (OC-3)
1999	2.5 Gbps (OC-48)
2002	10 Gbps (OC-192)
2008	40 Gbps
2012	100 Gbps per wavelength
2020	400 Gbps per wavelength
2024	800 Gbps per wavelength, with 1.6 Tbps emerging

The transition from OC-3 to 100G represented a 645x increase in per-link capacity over roughly 15 years, while the cost per bit dropped by similar magnitudes.

Core Routing: Where Decisions Are Made

At the heart of the backbone are core routers — extremely high-performance devices that make billions of routing decisions per second. Unlike the home router sitting in your living room, a backbone router might handle 100-400 Tbps of aggregate throughput.

Major backbone router platforms include: - Cisco ASR 9000 / CRS-X — Common in carrier networks - Juniper PTX series — Used by many Tier 1 operators - Nokia (Alcatel-Lucent) 7950 XRS — High-capacity chassis systems - Arista 7800R3 — Increasingly used in hyperscaler backbones

These routers run OSPF or IS-IS for intra-network routing (to find the best path within the carrier's own network) and BGP for inter-network routing (to exchange routes with other carriers).

Traffic Engineering

Modern backbone operators use MPLS (Multiprotocol Label Switching) with Traffic Engineering (TE) extensions to direct traffic along specific paths, avoiding congested links and distributing load. Instead of always following the shortest BGP path, MPLS-TE can route traffic around congestion, ensure bandwidth guarantees for premium services, and maintain fast failover paths.

More recently, Segment Routing (SR) is replacing classical MPLS-TE with a simpler model where the source specifies the path through the network by encoding a stack of instructions in packet headers.

Peering and Transit: The Economics of the Backbone

The backbone's physical infrastructure is only half the picture. The other half is the business relationships that determine how traffic flows and who pays whom.

Transit

Transit is a paid relationship where a customer pays a provider for access to the full internet. The customer's traffic enters the provider's network and is carried to any destination the provider can reach.

Transit prices have fallen dramatically: - 2000: ~$1,000 per Mbps/month at major PoPs - 2010: ~$10 per Mbps/month - 2024: ~$0.10-1.00 per Mbps/month at major IXPs, depending on volume

Peering

Peering is a settlement-free (or paid-peering) arrangement between two networks of roughly equal size to exchange traffic directly. Peering eliminates the need to pay a transit provider for traffic exchanged between the two networks.

Peering happens in two ways:

Private Network Interconnect (PNI) — Two networks physically connect their routers in a co-location facility and exchange traffic over a dedicated link. This is the highest-bandwidth, lowest-latency form of peering.

Internet Exchange Point (IXP) — A neutral facility where many networks connect to a shared switching fabric. Instead of building separate bilateral connections to hundreds of networks, a carrier connects once to the IXP and can peer with all other members on the same fabric.

Edge vs. Core Architecture

The backbone distinguishes between core and edge infrastructure:

Core routers sit deep inside the backbone, handling high-volume transit traffic between major PoPs. They prioritize raw throughput and forwarding speed. They typically do not perform packet inspection or complex policy enforcement.

Edge routers sit at the boundary between the backbone and customer or peer networks. They handle BGP sessions with customers and peers, enforce routing policies, apply QoS markings, and often perform rate limiting or filtering.

This separation allows the core to be optimized purely for speed while the edge handles the complexity of interconnection.

Resilience and Redundancy

Major backbone operators design for N+1 or N+2 redundancy on all critical links and nodes:

• Multiple fiber routes between major cities, following physically diverse paths (different conduits, different buildings, different geographic routes)
• Redundant routers at each PoP, with automatic failover
• Diverse power feeds, generators, and cooling systems in network facilities
• 24/7 network operations centers (NOCs) monitoring traffic and responding to failures

When a major fiber cut occurs — a backhoe slicing a buried cable or an anchor dragging across a submarine cable — backbone networks typically reroute traffic in seconds using pre-computed MPLS fast reroute paths.

The internet backbone's resilience comes not from any single indestructible component, but from the massive redundancy built into both the physical fiber routes and the logical routing relationships.