Memory-Restricted Routing with Tiled Map Data

I/O-Efficient Multi-Criteria Shortest Paths Query Processing on Large Graphs

“…Observe that with the expansion of Dijkstra's traversal process, the search frontier (i.e., the vertices waiting for traversal in the priority queue) enlarges exponentially in small-world graphs, while it becomes geographically large in road networks [6]. Moreover, the search frontier usually expands uniformly.…”

“…Moreover, the search frontier usually expands uniformly. As a result, there is a high probability that when we expand the search space in one partition G i , Dijkstra's algorithm may expand the search space in all other partitions along the search front before coming back to G i [6]. In this circumstance, G i may have to be evicted from memory, resulting in expansive reloads.…”

“…As a result, it hinders information sharing among the shortest path computations of different criteria. Secondly, since the expansion boundary grows rapidly and uniformly as the search expands [6], it reduces the overlapping of the required data of different criteria, thus greatly decreasing the probability of data sharing. Thirdly, since partitions usually organize the data to take advantage of its data locality [50], [51], [79], frequent evictions of underutilized partitions and massive redundant I/O operations may be involved due to the rapid growth of expansion boundaries and the limited memory.…”