When Efficiency Causes Gridlock

Why building more capacity can paradoxically make traffic and data networks slower.

Imagine a crowded city where everyone is simply trying to get home as fast as possible. Every driver makes a selfish choice to minimize their own travel time, completely unconcerned with the broader system impact. Mathematically, these individual choices converge on a Nash equilibrium that is globally suboptimal. When German mathematician Dietrich Braess published his paradox in 1968, he proved something counterintuitive: adding a brand-new road to a congested network can actually increase overall travel time.

This is not merely a theoretical quirk. Real-world observations have shown that closing specific roads can paradoxically improve traffic flow. When Seoul, South Korea, undertook the Cheonggyecheon restoration project, removing an elevated highway actually reduced overall congestion by altering commuter choices. The exact behavioral shifts of drivers remain difficult to predict due to changing psychological factors and variable demand, and some urban planners debate whether such improvements stem strictly from Braess's paradox or the broader reduction in capacity leading to traffic evaporation.

Yet the phenomenon extends far beyond asphalt. Computer and data networks experience the exact same vulnerability. When engineers add high-bandwidth links to a packet-switching network or a power grid, the new capacity can trigger a sudden rerouting of packets that creates severe bottlenecks and degrades overall throughput.

The root of the problem lies in a fundamental tension: maximizing local speed or node efficiency frequently conflicts with maximizing system-wide velocity. Local optima constantly impose negative externalities on the rest of the network. By optimizing relentlessly for the individual part, we routinely sabotage the whole.