Layers of Development

We all work at different levels of abstraction. Some of us work down around the machine level. Others work in a purely virtual environment with no visible connection at all to the machine. Maybe you've been lucky enough to work at multiple levels simultaneously. I personally have seen at least a little of everything from the physical theories of electronic components to high level frameworks in high level languages. I have some experience with embedded assembly and reprogrammable hardware. But I've also done server and client side scripting.

The more you're exposed to the easier it is to know on what level a given problem needs to be solved. Let's say you're writing a complex system that has a web interface and some really nasty back end algorithms. Once upon a time, you might have tried to do everything in a C-type algorithm. It wouldn't be much fun building a large server side script in C. And you might not even be able to solve your algorithms issues in C. What if it's too slow? What if even assembly is too slow?

These days, you would start by downloading a framework and stubbing out the web service. Then you might do some of the business logic in Java or C, depending on how much processing power it needs. If you have some really heavy algorithms, you might even do a special hardware implementation. Each part of the problem is solved at a different level. Knowing which tool to use is one the most important decisions you'll make. And this depends on processing requirements, number of acceptable mistakes, development time, number and skill level of developers, and so on.

Each level requires a different mode of thinking. You wouldn't worry about abstraction and code elegance on an FPGA. But you might spend a lot of time looking for an easy to implement algorithm that's still efficient. In Ruby, you shouldn't be worrying too much about efficiency. If you are, perhaps you're in the wrong language. You should be worried about removing duplicate code (DRY - Don't Repeat Yourself), elegance, and expressiveness.

I wrote before on the efficiency/productivity issue. It's really an issue of context. The closer you are to bits and bytes or electronics, the more efficiency probably matters. On the other end it shouldn't matter, or you've done something wrong. Usually, you will want to choose the tool that will allow you to complete the project the quickest possible but still be usable. And usable depends on the situation. In some situations, usable might mean it runs OK on a grid of a thousand computers. In other situations, usable might mean it runs in real time on a two dollar DSP.


We now have multicores, grid computing, and all these other wonderful technologies on the horizon such as quantum computers, optical, chemical, biological, etc. Each computer and architecture has specific strengths. Languages that bridge the gap between the high level and the low level will become very important. Thus you can program at a very high level and allow the language to determine or at least help to execute the code in the most efficient way on the best hardware.

Take multicores and grid computing. A language which makes concurrent programming easy will be of more use than a lower level, faster language. If you can program at a very high level without having to worry too much about what's executing where and when, you have a lot of power. The program might run on a single chip, or maybe two, or maybe it's running on a thousand quad-core computers.


home