The Iron Yuppie

The New Yorker: Life in the Slow Lane

I love this article... as you may know, I'm a long time fan of understanding traffic; it's such a fascinating system. The article covers exactly how difficult it is to manage the traffic in New York. Why not build more roads? I'm glad you asked! There's an amazing theory called Braess's Paradox which explains that, believe it or not, adding more roads may actually INCREASE average traffic time! Not because more cars are going, but because cars that previously took long roads with low time changes due to congestion now switch to shorter roads where congestion makes things much worse. Click through the link for a detailed explanation but imagine the following:

You have two ways home, each with two sections of road on your way home. One takes 25 minutes + 1 minute for every car on that road for the first section, then 1 minute + 5 minutes for every car on the road for the second section. The second is the exact reverse, with the first section taking 1 minute + 5 minutes for every car on the road, then 25 minutes + 1 minute for every car on the road. It looks a bit like this…



Average length of time to get home for 1 car = 32 minutes
Average length of time to get home for 9 cars = 80 minutes

So the city comes along and puts a short cut in between the two shortest roads, each which only take six minutes… and the short cut is so good it only takes one minute to cover no matter how many cars are on it.



Average length of time to get home for 1 car (taking just the shortest roads) = 12 minutes!

Good times, right? Wrong!

Average length of time to get home for 9 cars (taking just shortest roads) = 92 minutes… ACK!

Though you can end up load balancing for a while, where certain cars continue to take longer routes, it turns out when all the roads reach capacity, it’s likely that you’ve INCREASED the average drive home. This basically results from the fact that you’ve put more load on the less scalable roads where roads that get congested very quickly then can slow drive time even more than if you just took the long way home. Now imagine having to do a calculation like this for NYC where, rather than one or two routes home, you have a BILLION routes home. I’ll leave solving that problem as an exercise for the reader.

But wait, the article gets even better!

Just as the curve of maximum "throughput"—moving as many cars between two points on a road as efficiently as possible—reaches its peak, it abruptly falls off the cliff and is squashed flat against the baseline of the graph.

Traffic engineering is the science of maximizing throughput. What makes traffic jams hard to understand, at least within traditional traffic-engineering practice, is that they tend to occur around the time that the road is performing according to the engineers' peak specification. One important development in understanding this "nonlinear" phenomenon came in 1992, when Kai Nagel and Michael Schreckenberg, two physicists at the University of Cologne, in Germany, began to apply a computational technique known as "cellular automata" (or C.A.) to traffic. In a C.A. model, highway capacity is represented as a two-dimensional grid. Each cell in the grid has one of two "states": empty or occupied by a particle, which in this case is a car. Unlike traditional mathematical models used by traffic engineers, where it is assumed that all drivers are the same, in a C.A. model the particles can be assigned values intended to represent different types of drivers: fast drivers, slow drivers,
tailgaters, and lane changers can all be represented in the model. The result is virtual traffic.

Um, by better of course I mean more geeky. BUT boy is it geeky! I love this... where else can you find a system where one second before it starts failing it's operating at absolutely peak efficiency! And modeling traffic based on different driver agressiveness levels? Yummy! If you're designing SimCity 5, please build this in... I'll be indebted to you forever!