This is a terrific treatise on the problem of scaling, and the Divmod strategy, which I recommend to anyone looking for a clear picture in mind. It will help you make scalable systems where ever you are.
For me, Mantissa was more than an intellectual curiosity. I built my site on top of it. Though I'm now migrating to App Engine for other, pragmatic reasons (hosted environment, ease of deployment, Google's CDN, not having to do system maintenance, integration with Google's auth system, integration with Google services, BigTable, great prices, PR), Mantissa is a terrific place to get started to learn about the basics of scalability.
Once Divmod is able to get the rough spots ironed out, Mantissa could be a major competitor to App Engine. What they could provide in a hosted environment includes things Google has either ruled not to do, or haven't done yet (such as COMET).
Wow I'm sorry, but that was a pretty bad article, it seemed like the author knew little about the subject?
First of all, it said rails doesn't scale. This is so inaccurate on so many levels. Mainly because everything scales. Obviously for some situations languages like C will deal tons better - like google search. And some languages will be able to support more connections per server (again, C and even Ruby without the rails framework on top). But at the end of the day - Ruby on Rails will scale like any other language/framework, just sort out the database side to shard/slave and your away.
Oh and it compared Ruby on Rails to the Google App Engine. Two totally different things.
Just doesn't seem like a well thought out article to me?
Early on, the author says: "If you write an application with Rails, you probably have to write a whole bunch of new code, or at least change around all of your old code, in order to get it to run on multiple computers." This is absolutely true. Rails "scales" by forcing programmers to do a whole lot of extra work (yes, sharding and secondary databases and lots of other things) which isn't particularly natural in the framework.
As a programmer, I'm not particularly thrilled to have to set up and maintain a memcached cluster and then write code to make sure my object reads pass through the caching layer. It's more work. It's more fragile. It probably forced me to revisit and modify working code.
The author wants to build a framework where scalable code is the default and natural way to write applications. Seems pretty well-thought out. He touches on many of the common pain points.
I think half the people who say "rails doesn't scale" are idiots and the rest are making fun of the first half for trying to reduce the issue to such a glib statement.
This is too classic not to be quoted for tl;dr visitors.