Akamai Paper Review

From: Ali Akhavan <akhavan_REMOVE_THIS_FROM_EMAIL_FIRST_at_cs.toronto.edu>
Date: Thu, 6 Oct 2005 19:17:59 -0400

 
Akamai Paper Review - Ali Akhavan Bitaghsir

The overwhelming traffic load recently imposed on the web servers is the motivation of Akamai to widely spread the content delivery servers across Internet. changing a centralized or a semi-centralized system into a distributed one has many advantages and will bring up many challenges as well most of them addressed in the paper.

Akamai has exploited the real potential of the web, namely distributing the delivery servers across it; for dealing with high request loads, the existing web servers have tried a number of solutions. The based solution is first to make a cluster of server (instead of one) and in more advanced solutions, multi-homing is used for adding reliability beside performance to the end-users. So, why didn't they think to the Akamai solution: because widely distributing the servers across the earth will incur a very high administrative cost, unless you have many many customers. This amount of revenue for compensating the administrative cost may not be brought to the company unless it provides necessary infrastructure for B2B relation (like the offline and online content delivery monitoring component at Akamai). From another point of view, we can say that Akamai is a refined Internet in the sense that it makes use of more exact algorithms for controlling the flow of traffic among its nodes for performing load balancing; such a fine-grained monitoring is possible only by having a bounded number of servers. Therefore, if Akamai scales more and more, whether it fails to maintain its quality of service across its many thousands of servers, or it will replace the current infrastructure of the Internet, i.e. all vendors will use its infrastructure to deploy a web server. (As the portion of its servers to the total amount of servers become higher, then Internet servers will become a minority to Akamai servers).

One point worth of noting is the authors' opinion on the drawbacks of the other solutions, namely clustering and multi-homing and claiming that Akamai has resolved them. I am actually against one of them : The difficulty of scaling clusters to thousand of servers. I think, the difficulty of having thousands of servers widely across the world is more than having thousands of servers in one place. However, Akamai's success is due a trade-off it is involved in, which was mentioned in the previous paragraph.
Received on Thu Oct 06 2005 - 19:17:37 EDT

This archive was generated by hypermail 2.2.0 : Thu Oct 06 2005 - 19:17:38 EDT