The Science and Art of Code Structure

Jan 12, 2019
Featured Image

Clearly, I should have been in game development. I've been coding since the third grade so software was always in the cards for me. I have a degree in Mechanical Engineering so I have the physics chops to handle pretty much any kind of real-world simulation from projectile motion and particle physics to fluid dynamics and even heat transfer. My masters degree is in Industrial Engineering so I know about things like optimization and simulation that are helpful for AI (among other things). And games are fun!

What I realized recently, is that writing "good" code is an Industrial Engineering optimization problem. You see, optimization problems deal with multiple dimensions with often multiple optimal solutions. Note the use of the word "optimal" and not "best". Optimization problems often have multiple variables and an astronomical number of potential solutions. So when you are optimizing code, what is your goal? Maximize speed of execution? Minimize size of compiled code? Minimize bandwidth used? Minimize processing power used? Minimize total operating cost? Maximize test code coverage? Why not all of them? Well...I'll tell you why.

One of the most famous examples of an optimization problem and how mind-bogglingly many solutions they could have is the "traveling salesman problem". The traveling salesman must visit some number of potential clients and then return home. In the simplest version, he is flying from city to city and the only concern is the order of the flights to the various airports. A much more practical and complex version is a package delivery company with a fleet of trucks needing to deliver packages over routes through a complex network of highways and city streets with variable traffic, one way streets, traffic lights, etc. Every variable multiplies the number of solutions by the number of possible options for that variable. The number of possible solutions very quickly gets large – like more than the total number of atoms in the known universe large. What hope do we have of solving these problems? Luckily, we have pretty cool brains.

Malcolm Gladwell wrote a book titled Blink: The Power of Thinking Without Thinking. It's a great book and I highly recommend reading it, but the gist is our brains are able to take in a vast array of information and come to a conclusion (an "optimal solution") without consciously thinking about it. This ability to organically problem solve was also recently discovered in single cell organisms. This is why we are pretty good at coming up with decent solutions to these problems without knowing anything about linear algebra. Our brain operates on "heuristics" that our life experiences have allowed our brains accept without having to think about them. In the world of software development, the use of these heuristics are the "art" and allow us to optimize variables that are extremely more challenging and seemingly abstract such as "maximize code reuse" and "maximize maintainability" and, as I like to say, "minimize astonishment".

Another important consideration in optimization is boundaries. For example, you may be able to purchase a certain amount of computing resources for a certain cost. Additional resources add more cost. This means that you don't necessarily need to use an absolute minimum number of resources, you just need to make sure that the amount of resources used doesn't trip the threshold or that if it does, you have a source of revenue to cover the added cost.

In future posts, I will dive into more detail on the day-to-day practical advise on optimizing your code, but first I wanted to take some time to lay some groundwork. In the early days of computers, you were greatly constrained by memory and processing power. You had a limited number of pixels that could be rendered on the screen in a limited number of colors. You could only store so many bytes of code. You were limited to a small number of operations the processor performed. You were limited in the language you could code in and the structure of that language. The optimization required was much more science with a small number of options for each variable. Now, the options are astronomical. What language do you use? What platforms or devices run that code? What communication channels does it leverage? How many millions of colors and pixels does it render? Sure, it is still important to minimize operating cost (storage, CPU cycles, bandwidth, etc.), but you can also get access to a significant amount of resources for little or no cost. This allows you to focus on the art – maintainability and reuse – first, and then solve the challenging scientific problems later (or not at all).