Review - Globally Distributed Content Delivery

Akamai provides scalability and reliability to an online service without the
need for over provisioning. Typical mechanisms for handling high service
load on the Internet include caching, clustering and multihoming. Each of
these requires that an organization pay for services and system that are not
operating at full capacity. This translates to lost revenue. Utilizing
shared resource on demand in a cost effective manner is a complex problem.
Akamai provides "capacity on demand", which helps to mitigate flash crowds
and avoids a single point of failure, while allowing the organization to
maintain control of their data.

By working closely with the customer, and using advanced internetworking
constructs, Akamai is able to offer a highly scalable, flexible and fault
tolerant service. Dynamic application content can be modularized and
assembled on edge servers. Streaming content is supported, where lightly
loaded nodes are able to take over failed node operations in a manner
transparent to the user. And, user authentication tokens can be negotiated
with origin servers. These, and other, services are realized over Akamai's
flexible server to client mapping schemes. Here, DNS redirection is the
highest level of resource allocation and load balancing. Akamai agents also
communicate with BGP routers to gain an overview of the network and detect
most efficient paths.

The most impressive aspect of the Akamai system is its integrated real-time
monitoring abilities. While the Internet is inherently difficult to measure,
Akamai has incorporated measurement as an essential system characteristic.
Agents simulate end users, which provides measurement of failure rates and
aids in early detection of faults. Live traffic can be duplicated and used
for online testing, or used to test new software before deploying upgrades.
These monitoring constructs allow Akamai to react quickly to changes and
provide their customers with real-time statistics.

The Akamai system attempts to solve a critical problem in the business
space. On one hand, over provisioning is not cost effective and on the
other, fault tolerance, scalability and capacity are desired in order to
retain customer satisfaction. Akamai provides a solution that delivers
capacity on demand while overcoming interesting and exciting technical
challenges. Although this was not a scientific paper, providing no
measurement results or implementation details, it does an excelent job of
outlining some of the key constructs in a successful distributed system.
Received on Thu Oct 06 2005 - 01:45:36 EDT