Eric.
Thanks for your prompt response. You're right, my question was too broad. Let me try another approach.
AOL has over 20,000 servers and about 25,000,000 users surfing, transacting, etc. Assuming that AOL remains with Sun/Network architecture, they drop in a server for every 1,250 users. If AOL ever captured 100 million US citizens, they would have to have 5X the hardware! Does 100,000 servers seem possible to you? Of course, if they captured an even larger segment in India and China, say, another 900 million people in addition to US subscribers (1 billion citizens total world-wide), they would need 800,000 servers! Does this seem even remotely possible to you?
My point is, I don't see how ANY single hardware/software entity can capture hundreds of millions of users, let alone a billion or two billion or three billion. Ergo, peer-to-peer.
You mentioned that you liked inexpensive, distributed computing over big iron. If a Napster clone, say, Flycode or Scour, gets its act together with respect to "copyright friendliness" and security, would this be a platform to love? Please elaborate. In one swoop, a well-built peer-to-peer eliminates 99% of the scalability issues. True or not?
Jud Jud, It’s funny you mention peer-to-peer. Afterall, isn’t the entire Internet one large peer-to-peer network? Let’s continue using AOL as a case in point. You register skepticism regarding AOL’s ability to scale up to 1 Billion users. I would have to disagree. I believe it is not only possible but not even tremendously difficult for AOL to scale to 1 Billion users simply by extending their current platform over time (Note: I think this will be over a number of years and through at least 3 or 4 generations of hardware/software improvements). Let me give you a for instance why this isn’t the problem you believe it is. One of the biggest problems AOL has is storing all of this data. AOL is one of Network Appliance’s biggest customers. And just recently Network Appliance announced first shipments of their new line of Filers, the 840c. This Filer supports 12 TB of storage (up from the previous generation of approximately 1.5 TB, an 8x improvement), 3.4X throughput performance increase to 25,900 SFS ops/sec @ <1.6ms overall response time (50% improvement) and a 3.9X net bench throughput performance increase to 987 Mbps.This is a huge evolutionary step. Potentially, by tearing out the current AOL architecture and replacing it with these systems, AOL could realize more than a 4X capacity increase and easily hit the 100 Million user level (from a storage perspective only). At the same time, Sun has just announced the first member of their new UltraSparc III Sun Fire Family of Servers, the Sun Fire 280R. Again this is another evolutionary step that should keep AOL on the crest of the technology wave necessary to manage their growth. At the same time, AOL engineers are continually rewriting and optimizing the AOL code base to improve performance. You contend that in order for AOL to support 100 Million US users, they would need something like five times the hardware. I would suggest that they might need twice the hardware, but certainly not five times the hardware. Since all of the new hardware they would purchase would be high quality fast and modern hardware capable of supporting on the average significantly more customers than the current batch of hardware they currently use. At the same time, they will be replacing older hardware that is more costly to maintain and support than the incremental benefit the older hardware provides. Will all of these measures mean that AOL could reduce the number of servers and continue growing at the same rate? No, probably not. However, it does mean that they could continue to build out their infrastructure with the newest servers and tools and still maintain control over their environment. I also think that an entity like AOL will continue to build out centers in other countries in order to help them manage the growth better and maintain services closer to their customers (reducing network latency and long haul bandwidth requirements). As for peer-to-peer in the sense that everyone’s computer in the world could potentially be a server, I agree it is possible. The Napster induced peer-to-peer craze does not appear to be waning; however, I would caution you against buying into the idea of placing your home computer on the Internet as a server for all. What you will end up with is a huge security risk with no real benefit to yourself by opening up your home computer to every Tom, Dick and Sally on the Internet. Thanks again for your note. This has been a very interesting discussion so far (I hope it continues) and I would like to open it up at this point to everyone on IThell.com. Please send your responses to letters@IThell.com. Eric Svetcov Home |