The urge to scale
I guess being a dot-com CTO is in my blood. I like to think through various architectures for managing groups of websites. You need to lock down a model for scaling early or you face big problems if you ever need to handle large amounts of traffic. The real key is a logical architecture for domain names. For example, if I thought I was going to serve a lot of podcasts, I would create something like data.darwinianweb.com or podcasts.darwinianweb.com. That would allow me to move that part of my content where it could be best and most cheaply served.
Right now I have darwinianweb.com to handle this main blog where I plan on covering general issues on the changing form of the Internet. I also have ruby.darwinianweb.com, which is a blog that allows me to go into as much depth as I want about learning the Ruby programming language.
I don't want to have too many subdomains, categorization can be handled more easily and on a larger scle with tags, which I am working on adding. At the same time, a separate domain creates more of a distinct place or channel of thought for the user. People automatically switch contexts when they change to a new site, just like a new TV channel.
I plan on having only a few more content subdomains, such as ajax.darwinianweb.com, and xml.darwinianweb.com. Programming languages or standards like XML are so broad and have so many supplementary tools and resources that they work better in their own site or subsite.
I'll also be creating separate domains for exchanging data with other servers. I don't know what will happen with my API experiment, or if that will become a target for abuse, so I'll also create api.darwinianweb.com to serve API calls. It isn't a matter of large amounts of traffic. I want to be able to shut down the API server easily. Of course, that brings up the issue of dependency on critical servers in a distributed environment called for by Web 2.0.
One solution, which also comes easily in an XML/RSS based communication model, is cache the most recent messages as text files, so the most recent result of an API call can be reused instead of calling the API again.
These issues will be played out on a much larger scale throughout the web. Chains of API dependencies will play interesting roles in the future.
