Aehso’s Output

Experience with some Principles for Building an Internet-Scale Reliable System

For distributed system engineers only, three Akamai engineers have submitted a very interesting paper to the World ‘05 Second Workshop on Real, Large Distributed Systems. You don’t have to be building a large distributed system for many of the principles they employ to apply though - their approach addresses a couple (but not all) of the Eight Fallacies of Distributed Computing. They do employ some very simple but effective tricks to increase availability - reducing the TTL of just 20 seconds for IP mappings in their LLNS servers is a good example.

Akamai is a company I’ve always been fascinated with. They managed to quietly develop a truly transparent, global and reasonably intelligent content delivery platform that is second to none. For example, if you’ve ever bought music from the iTunes Music Store then you’ve used Akamai’s servers. Most global high-bandwidth sites like Yahoo!, Google and Microsoft/MSN use Akamai to distribute their brunt of their HTTP content. You wouldn’t know this from Akamai’s customer testimonial page but last years Akamai DNS outage highlighted who their customers are and how much they depend on Akamai.

No comments yet.

Leave a comment