The core of this issue is that we’re in a **PARALLEL COMPUTING ENVIRONMENT**
That’s really the issue. You’re planning on using multiple CPUs to do the work, and the architecture bottlenecks at your ability to get the centralized components of the architecture to behave.
Gee… didn’t the supercomputer guys have this problem about 30 years ago?
A single system, serving hundreds or tens of thousands of concurrent users is going to have these problems. The more interactive it becomes, the more **LIKE SOFTWARE** your web site is, the worse the problem will be.
You think you have problems? Check out:
and imagine this problem in the context of sixty million people hitting your system in a three hour window to try and figure out how to save themselves from a disaster.
Can it be done? Sure. But the STORM botnet is a your model, not your current web applications framework.
Start thinking of swarms of disposable entities serving your web sites. It’s not about taking a single webserver app and scaling it no no no. It’s about writing parallel supercomputer cluster software that happens to be user-interactive rather than number crunching.
Nodes that carry data. Migration over the network to spare compute resources. Local session data that’s spooled to disk replicated on the fly rather than databases.
SQL *on the node* if you need it (think SQLite) rather than an Oracle install.
It’s not easy: people in the supercomputer world have been worrying about this since the 1980s, maybe earlier. But that’s what you’re up against here: supercomputer sized computations with irregular datasets and complex interdependencies.
If you haven’t thought about this stuff, you need to go back to the Transputer papers and the Occam programming language. Occam is really the best single source for clear thinking about parallel computation as a fundamental part of software design that I am aware of, and the Transputer is an idea who’s time is not yet come – first we need parallel programmers, *then* parallel computers.
Once you’ve got your head around that – and things like the “Crazy Postman” algorithm and why they worked in a transputer environment – then you’re ready to get down to writing code for these massive new generation applications.