Inferensys

Glossary

Content Delivery Network (CDN)

A geographically distributed network of proxy servers and their data centers that caches content close to end-users, reducing latency and improving load times for static assets.
Data scientist building training data pipeline on laptop, data preprocessing visible, technical workspace.
DISTRIBUTED INFRASTRUCTURE

What is a Content Delivery Network (CDN)?

A Content Delivery Network (CDN) is a geographically distributed network of proxy servers and their data centers designed to cache content close to end-users, reducing latency and improving load times for static assets.

A Content Delivery Network (CDN) is a geographically distributed group of servers that work together to provide fast delivery of Internet content. By caching static assets—such as HTML pages, JavaScript files, stylesheets, images, and videos—on edge servers located closer to end-users, a CDN minimizes the physical distance data must travel, drastically reducing latency and improving page load performance.

When a user requests content, the CDN routes the request to the nearest Point of Presence (PoP). If the content is cached, it is served instantly; if not, the CDN retrieves it from the origin server, caches it, and delivers it to subsequent users. This architecture also provides load balancing, absorbs traffic spikes, and mitigates Distributed Denial-of-Service (DDoS) attacks by distributing the request load across its global infrastructure.

CORE CAPABILITIES

Key Features of a CDN

A Content Delivery Network is more than just a cache. It is a globally distributed platform designed to optimize every stage of content delivery, from origin shielding to dynamic request routing.

01

Edge Caching

The foundational mechanism of a CDN. Static assets—such as HTML files, images, CSS, and JavaScript—are stored on proxy servers at the network's edge. When a user requests a resource, the CDN serves it from the geographically nearest Point of Presence (PoP) rather than the origin server.

  • Cache Hit Ratio: A critical performance metric measuring the percentage of requests served directly from cache.
  • Time-to-Live (TTL): Defines how long an object remains fresh in cache before the CDN revalidates with the origin.
  • Cache Purging: The ability to instantly invalidate cached content globally, essential for updating time-sensitive information.
95-99%
Target Cache Hit Ratio
02

Origin Shielding

An intermediate caching layer that sits between the edge servers and the origin infrastructure. Instead of every edge PoP requesting the same uncached object from the origin, a single shield node makes the request, caches the result, and distributes it to all edge servers.

  • Origin Offload: Reduces requests hitting the origin server by up to 99%, protecting it from traffic spikes.
  • Consolidated Requests: Prevents a thundering herd problem where thousands of edge servers simultaneously request the same expired object.
03

Anycast DNS Routing

The network addressing technique that makes CDN geography transparent. A single IP address is announced from multiple locations simultaneously. The Internet's Border Gateway Protocol (BGP) automatically routes the user's request to the nearest available PoP based on network topology, not just physical distance.

  • Latency-Based Routing: Directs traffic to the PoP with the lowest round-trip time.
  • Failover: If a PoP becomes unavailable, BGP automatically withdraws its route, redirecting traffic to the next closest node without manual intervention.
04

Dynamic Content Acceleration

Optimization techniques for non-cacheable, personalized content. Unlike static assets, dynamic requests must reach the origin, but the CDN accelerates the transit. This is achieved through route optimization, persistent connections, and protocol enhancements.

  • TCP Pre-Pooling: Maintains warm, pre-established connections to the origin to eliminate the three-way handshake overhead.
  • Route Optimization: Uses real-time internet telemetry to bypass congested peering points and find the fastest path to the origin.
  • Protocol Upgrades: Terminates HTTP/1.1 at the edge and upgrades the origin connection to multiplexed HTTP/2 or HTTP/3.
05

Edge Compute (Serverless)

A runtime environment distributed across the CDN's edge nodes that allows developers to execute custom logic at the point of presence. This moves computation from a centralized cloud region to the network's edge, enabling sub-millisecond cold starts.

  • Request Modification: Rewrite headers, URLs, and cookies before the request reaches the origin.
  • A/B Testing at the Edge: Execute traffic splitting logic without adding latency to the user.
  • Authentication: Validate JSON Web Tokens (JWTs) at the edge, blocking unauthorized requests before they consume origin resources.
06

Image and Video Optimization

Real-time media transformation performed at the edge, eliminating the need for pre-generated variants. A single high-resolution source image can be dynamically resized, cropped, and converted to modern formats based on the requesting device's capabilities.

  • Format Negotiation: Automatically serves WebP or AVIF to supporting browsers while falling back to JPEG for legacy clients.
  • Responsive Images: Generates exact dimensions on-the-fly via URL parameters, reducing bytes transferred.
  • Quality Compression: Applies perceptually lossless compression to reduce file size without visible degradation.
CDN ESSENTIALS

Frequently Asked Questions

Clear, technically precise answers to the most common questions about Content Delivery Networks, their architecture, and their role in modern web infrastructure.

A Content Delivery Network (CDN) is a geographically distributed network of proxy servers and their data centers designed to cache content close to end-users, reducing latency and improving load times for static assets. When a user requests a resource, the CDN's request routing system directs them to the nearest edge server rather than the origin server. If the edge server has a cached copy of the content, it serves it immediately; if not, it retrieves it from the origin, caches it, and serves it to subsequent users. This architecture leverages Anycast DNS to route traffic to the topologically nearest Point of Presence (PoP), minimizing round-trip time (RTT) and offloading bandwidth from the origin infrastructure.

INFRASTRUCTURE COMPARISON

CDN vs. Traditional Web Hosting

A technical comparison of content delivery architectures for serving static and dynamic assets to a global user base.

FeatureContent Delivery NetworkTraditional Single-Origin HostingEdge Computing Platform

Architecture Model

Globally distributed proxy servers caching content at multiple Points of Presence (PoPs)

Centralized origin server in a single data center serving all requests

Distributed compute and storage at the network edge with serverless execution

Latency Profile

< 20ms for cached assets from nearest PoP

50-300ms+ depending on geographic distance from origin

< 10ms for edge-executed logic

Origin Server Load

60-95% reduction in requests to origin

100% of requests hit the origin server

70-99% reduction; dynamic logic runs at edge

DDoS Mitigation

TLS Termination

At edge PoP, reducing handshake latency by 40-60%

At origin server only

At edge node with automatic certificate management

Cache Invalidation Speed

Instant via API purge; sub-second global propagation

Not applicable

Instant via API; supports surrogate key tagging

Dynamic Content Handling

Cache bypass rules forward to origin; limited edge logic

Native server-side processing for all requests

Full serverless compute at edge; database queries run locally

Global Load Balancing

Anycast DNS routes users to nearest healthy PoP automatically

Requires separate third-party DNS load balancing solution

Built-in anycast with health-aware routing and failover

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.