Eric,
Enjoyed your comments regarding Sun (ZDNet Talkback, "UltraSparc is Only 1/2 the Story"). If you have a free minute, would you mind answering a question?
If YOU had very deep pockets, were starting a global Web site that had to scale to the ends of the galaxy in terms of flawless consumer transactions (potentially everyone on the planet), what would YOUR system look like: servers or mainframes? And why?
As I understand it, AOL has 20,000 Sun servers and about twenty-some million customers; that's one hell of a server farm. I do not see how AOL, or anyone for that matter, can keep adding servers in an endless procession without mangement and/or equipment meltdown. True or not?
If YOU wanted to scale to everyone on the planet, would mainframes be the better solution, HP in particular? (I mention HP because I'm considering them--and IBM--as hardware providers.)
I'm forever asking this question because I'm forever getting contradictory information. One would think by now that the issue of huge, clustered farms vs clustered mainframes would be settled. Both entities have their take--redundancy, management costs, etc.--and it's very dificult to sort out. I would very much appreciate your view.
In addition to costs, ease of administration, and flawless operation issues, bandwidth choke points also loom as a potential problem. Does this mean that the hardware should be geographically distributed, too?
Perhaps my question should be: Is there ANY design/model that would handle--not just theoretically--literally hundreds of millions of customers simultaneously?
Jud
Jud, Unfortunately, the scope of your question is too broad. You say you want a web site that could scale endlessly. What exactly are you going to do with this website? Is it just static web pages? Is it for e-commerce? What applications, application environment, database etc. do you intend to use? There really is no way to answer the question as posed. There are advantages to just about any solution you could think of. For scalability there are a series of issues: Networking issuesHardware issuesApplication issuesDatabase issues Personnel issuesCost Let me just give a for instance on Personnel issues because frequently they are the most difficult to handle....imagine you decide you are going to go with an HP Mainframe solution, but you find that you can't hire enough personnel to properly staff the operation (for whatever reason). Now, that HP Mainframe solution is no longer the best solution because you can't get the right people to operate it. We need some constraints on this issue....before going further....also, if you don't mind, I would like to open it up on the IThell.com website. Let folks there give you feedback....maybe someone would come up with something that strikes a chord in your mind. You did ask specifically if there is a way to handle hundreds of millions of customers simultaneously....there is, however, most companies would scale up gradually and change their architecture many times on the way. For instance, AOL started on Stratus systems....that's a long way from their current Sun/Network Appliance hardware architecture. Also, in general I'm a guy who likes inexpensive distributed computing over big iron.....but that also may be because I really know that best. By the way, if you get funding for this project, I might be interested in doing some consulting.....sounds like a fairly significant idea if you want to scale to the world :) Sorry I can't be more helpful..... Eric Svetcov