The Akamai Network: A Platform for High-Performance Internet Applications
📜 Abstract
Comprising more than 61,000 servers located across nearly 1,000 networks in 70 countries worldwide, the Akamai platform delivers hundreds of billions of Internet interactions daily, helping thousands of enterprises boost the performance and reliability of their Internet applications. In this paper, we give an overview of the components and capabilities of this large-scale distributed computing platform, and offer some insight into its architecture, design principles, operation, and management.
✨ Summary
Paper summary
The paper presents Akamai as a large-scale distributed computing platform designed to improve the performance, reliability, scalability, and security of Internet applications without requiring changes to the underlying Internet or end-user software. Its central architectural idea is a virtual delivery network implemented as a software layer over widely distributed servers positioned close to users and, where possible, close to access networks.
The platform combines several major capabilities: edge servers for request processing and caching; a mapping system that selects servers using geographic, topological, performance, load, and health information; transport systems for reliable communication between origins, edge clusters, and intermediate nodes; communications and control infrastructure; large-scale monitoring and data-analysis systems; globally distributed DNS; storage; and a customer management portal. A recurring design assumption is that failures are normal. The system therefore emphasizes redundancy, automatic recovery, fault isolation, decentralized coordination, staged configuration changes, and minimal dependence on human operators.
For static content and media, the paper describes edge caching, tiered distribution, and a live-streaming overlay. Tiered distribution reduces origin load by routing cache misses through parent clusters, while live streams are replicated through reflector nodes and delivered over alternate paths. Multiple paths, packet-recovery techniques, and prebursting are used to reduce loss and startup delay. These mechanisms support the paper’s argument that a highly distributed architecture is better suited than a small number of centralized data centers for large global audiences and high-bitrate video.
For dynamic applications, Akamai uses its distributed servers as a performance-optimized overlay. Path selection, persistent connections, transport-protocol tuning, intelligent retransmission, compression, and prefetching reduce the effects of latency, packet loss, inefficient BGP routing, and middle-mile congestion. The platform also supports EdgeComputing, allowing selected application components to execute near users while retaining centralized origin services where necessary.
Influence on research and industry
The paper helped document and consolidate the architectural principles behind commercial content delivery networks, including edge caching, DNS-based server selection, overlay routing, distributed fault tolerance, and edge execution. Later research on information-centric networking, content placement, cache allocation, fog computing, and edge computing cites the paper as an example of large-scale content-delivery infrastructure. (sciencedirect.com)
Its treatment of live-streaming overlays also connects directly to subsequent algorithmic research. A later paper on overlay-network construction states that its algorithms include networks used by Akamai for global live-media delivery and evaluates them using traces derived from Akamai’s live-streaming network. (arxiv.org)
The paper continues to be used as an established systems reference in computer-networking and distributed-systems courses, and later Akamai technical publications cite it when describing the company’s research and delivery-network architecture. (courses.cs.washington.edu)