AccessMyLibrary provides FREE access to millions of articles from top publications available through your library.
Original Source: FD (FAIR DISCLOSURE) WIRE
. Sandy Smith, Akamai Technologies Inc., Director, IR . Tom Leighton, Akamai Technologies Inc., Founder, Chief Scientist . Mike Afergan, Akamai Technologies Inc., CTO . Harry Blount, Lehman Brothers, Analyst
. J.D. Sherman, Akamai Technologies Inc., CFO . Ray Conley, Palo Alto Investors, Analyst . Chuck Jones, Atlantic Trust, Analyst . Gina Sockolow, Colonial Sun, Analyst . Danny Joe, TIA Acram, Analyst
AKAM reported that highly distributed computing matters because the Internet users are highly distributed, and by having greater distribution, it has performance, reliability, scalability, and lower cost infrastructure. Co. is in 1,200 different physical locations and in over 900 different networks.
S1. Highly Distributed Computing (T.L.) 1. AKAM Platform: 1. Has over 25,000 servers. 2. AKAM is in 1,200 different physical locations and in over 900 different networks. 1. AKAM is in well over 1,200 locations in 650 cities and 69 countries.
3. AKAM's model is that wherever its customers' client is in the
world and however they connect to the Internet, Co. wants to
have its servers to be in that same city and in that last-mile
network. 4. Serving close to 200b hits per day, well over 100m streams, and over 3 petabytes of traffic. 5. Percent of Web traffic that Co. serves in any given day can be between 15-33%. 6. Four reasons why highly distributed computing matters:
1. Internet users who are trying to serve quickly and reliably
by themselves are highly distributed. 1. In many locations around the world. 2. By having greater distribution of its infrastructure, Co. can give greater performance and reliability to those distributed users of the Internet. 3. By having greater distribution, Co. can have greater scalability, which is particularly important as media and video moves into the Web in a major way. 4. By greater distribution, Co. has a lower cost infrastructure. 2. Highly Distributed Internet Users: 1. Where the users connect to the Internet and where the infrastructure, the origin servers, they all connect to an ISP. 1. These ISPs are typically different.
2. They are at the edge of the network. 3. One hasn't heard of their networks before. 2. Internet is composed of 13,000 different networks. 1. Largest network today controls a little over 8% of the access traffic on the Internet. 2. The next one is at 5%. 3. Taking top 30 networks in the world, collectively, they only get 50% of the traffic. 4. If one wants to get 90% of the access traffic, one got to go to 1,000 different networks.
5. AKAM is close to 1,000 networks and directly connected in last-mile to about 90% of the Internet. 6. There are 13,000 networks that are competing against each other. 3. Distance from typical end user depending on how many points of presence one has with the servers. 1. If one is point of presence or a single central location, avg. distance to an end user, if it's global operation, will be well over 3,000 miles. 2. If one is truly global and has a lot of activities in Asia, Europe and the US, maybe closer to 5,000 or 6,000 miles. 3. If one is more US-centric, maybe closer to 3,000 miles would be the typical avg. distance to an end user. 4. If one has a few dozen locations, typical distance to an end user is about 1,500 miles. 5. If one is in 1,000 locations like AKAM, typical distance to an end user is well under 250 miles today. 4. Economic Truths of the Internet: 1. Internet users themselves are highly distributed. 2. Capacity and bandwidth is not free and infinite. 1. Bandwidth prices have come down substantially over the last decade, and there has been a lot of growth in the
bandwidth and the Internet, but it still is expensive for end users and it is not [infinitive]. 3. Dollars to purchase Internet infrastructure that drives the business of the Internet flow in at the edge of the network where the end users are and where the origin infrastructures are. 1. They are buying bandwidth and access to the Internet. 4. Routing algorithms are not based on performance
considerations, they are based on economic considerations. 1. Don't take reliability, latency, and congestion into account. 2. Dictated by economic considerations between the various networks, which are competing with each other. 3. This results in very poor routes and reliability for the Internet. 4. Most typically occurs at the peering points between the networks, and there are no economic incentives to make those peering points become more robust or become better. 5. The economics work just the reverse. 3. Greater Distribution Means Greater Performance and Reliability: 1. Drives a lot of AKAM's business. 2. Latency considerations when trying to serve from a great distance to the end user. 1. A normal Internet routing creates paths that have high latency and packet loss because the paths go to multiple networks and as they get clogged in peering points, one introduces latency and packets of loss. 2. AKAM on the other hand has a built on an overlay network on top of the Internet that overcomes the economic difficulties, challenges, and performance issues between 13,000 networks that make up the Internet.
3. Co. finds a good path that it can use, but nobody else can
use through the Internet that makes it fast across the long distances, if one has to reach an origin infrastructure, and when serving near to the end user, makes it even faster. 4. This translates immediately into topline revenue increases for customers. 5. Ross-Simons, an e-commerce site, when they turned on AKAM, the site became over three times faster for end users. 1. Recorded $20m increase in annual revenues directly attributed to the performance improvement. 2. At the same time, AKAM saved them $300,000 on infrastructure savings.
3. Latency determines the throughput that's available. 1. Since AKAM is serving it locally, the round trip is only 10 milliseconds instead of 100 milliseconds. 2. If serving from a single location and 3,000 miles away, one is going to get less than 500 kilobytes a second. 3. If serving from a few dozen locations, one is going to get less than a megabyte a second to the typical end user.
4. If one is in over 1,000 locations, one can get more than 10 megabytes a second to the typical user.
S2. Highly Distributed Computing (M.A.) 1. Greater Distribution Means Greater Scalability: 1. One of the other key challenges is having the scalability and the ability to be able to serve those large volumes of content at the rate. 2. Internet is made up of these various networks who connect at these peering points.
3. Given the economic structure of the Internet, often needs
peering points what's called the settlement-free, that is no
one is paying each other for the connectivity through that
peering point. 1. While they have some benefits to the Internet, one of the fundamental challenge is there is no incentive for these
networks to be building out the capacity, building out the scale at these peering points. 2. They become fundamental bottlenecks to traffic, to content moving through the network, and it becomes a bigger and bigger problem. 4. AKAM places its servers on the other side of the peering point. 1. Once moving to the other side, there is a fundamental difference between being close to the ISP and being inside the ISP. 2. In AKAM …