Dynamic load distribution turns where a route is computed into a runtime decision, and that changes what you have to protect.
Every navigation feature hides a decision: where does the computation run? Put all of it on the phone and weak devices suffer. Put all of it on the server and the network becomes the bottleneck, along with the bill.
The client-server approach in this study treats that decision as something to make continuously, rather than once at build time.
The first signal is calculation complexity. A short route across a small graph is cheap. A long one across a dense network is not, and the cost shifts as the graph changes.
The second signal is device capability. Because memory capacity turned out to be the strongest influence on Dijkstra's calculation time, the client's hardware profile is worth measuring rather than assuming.
Moving work to the server buys consistent performance on cheap hardware. It also buys a dependency. The route search now needs a working connection, and the server takes on load from every client that cannot cope on its own.
The paper is careful about that second half. Adjusting the Dijkstra calculations in real time is what keeps the arrangement from collapsing when demand spikes. Prevent the server from overloading, and the experience stays steady for everyone.
The stated goal is a waiting time that never exceeds one second. Working backwards from that is more useful than optimizing an average. If the target is a wait no user notices, the interesting question is not which device is fastest. It is which devices miss the target, and what the system does about them.
That framing travels well. Any feature that pairs a phone with a backend has a class of devices that will be too slow for it, and a decision to make about whether to help them or leave them behind.
Justico et al. (2023). Dynamic Load Distribution of Shortest-Path Finding in Client-Server Public Transit System. 2023 International Conference on Modeling and E-Information Research, Artificial Learning and Digital Applications (ICMERALDA).
DOI10.1109/ICMERALDA60125.2023.10458151 (opens in a new tab)
