Choosing a routing method is a trade off between time, memory, and the phones your users actually carry.
Choosing an algorithm for a product is rarely about which one is mathematically better. It is about which cost you are willing to pay, on the hardware your users actually carry.
A 2023 study compared two ways of recommending public transit routes: the long established Dijkstra algorithm and ant colony optimization, which mimics how ants find short paths. The goal was not to crown a winner, but to see whether the newer method could match the older one under the same conditions.
The team tracked computation time and memory consumption, including how much memory each method held at the start and at the end of a run. Those two numbers are usually what decides whether a feature feels smooth on a mid range phone.
Three habits from this study are worth copying into your own benchmarks.
The measurement discipline here is easy to underestimate. Running the same test many times protects against one off artifacts, and the researchers call out the garbage collector specifically. In a managed runtime, memory behaviour during a single run can be an accident of timing rather than a property of the algorithm.
For teams building transit features, the practical advice is to define what you are optimising for before writing the code. If memory is scarce, the classic algorithm is a safe default. If you are searching a very large network and can afford the footprint, the ant based approach deserves a test on your own data.
Abdul Jabbar and Aryo Pinandito (2023). Recommending Public Transit Route: Ant-Colony Optimization or Dijkstra? 2023 International Conference on Modeling and E-Information Research, Artificial Learning and Digital Applications.
DOI10.1109/ICMERALDA60125.2023.10458207 (opens in a new tab)
