Abstract
This paper describes a new method to compute isochrones in multimodal spatial networks, which aims at finding a good trade-off between memory usage and runtime. In the past, approaches based on Dijkstra's algorithm have been proposed. For small networks, the entire network is first loaded in main memory, where the network is expanded to determine the isochrone. For large networks that do not fit in main memory, approaches that load the network vertex-by-vertex during the expansion phase have been proposed. They keep the memory footprint minimal, but have to query the database for each node in the isochrone, which can be very time consuming. The method presented in this paper uses tiles (which are well known from interactive online maps) to realize chunk-loading of vertices by utilizing so-called tile regions. This approach significantly reduces the number of database requests, while keeping the memory usage low. Our method is able to compute isochrones even in large networks at a reasonable time. An experimental evaluation shows that the new algorithm clearly outperforms previous competitive approaches such as MINE and MINEX.