A Comparison Between Label-Setting and Label-Correcting Algorithms for Computing One-to-One Shortest Paths ∗

It has been a general belief that label-setting algorithms are better choices than label-correcting algorithms for computing the shortest path distance between a source node and a destination node (one-to-one) on a network. The reason is that iterations in label-setting algorithms can be terminated whenever the destination node is scanned and permanently labeled. In this paper, a fast-performing implementation of each type of algorithm is selected based on the results of a recent study (Zhan and Noon 1998) and compared for computing one-to-one shortest paths using 10 large real road networks. The selected label-setting algorithm is Dijkstra's algorithm implemented with approximate buckets (DIKBA) and the selected label-correcting algorithm is Pallottino's graph growth algorithm implemented with two queues (TWO-Q). It is concluded that, in some situations, the TWO-Q algorithm is a better choice for computing one-to-one shortest paths on real road networks.