Parallelized Proximity-Based Query Processing Methods for Road Networks

TL;DR

This paper introduces parallel methods for proximity-based query processing in road networks, enabling efficient distance and k-nearest pairs searches, with a novel graph partitioning heuristic for improved parallel computation.

Contribution

The paper presents a new parallel graph join paradigm for road networks and a tailored recursive bisection heuristic for effective graph partitioning.

Findings

Efficient parallel processing of proximity queries in road networks.

A novel heuristic for graph partitioning tailored to the problem.

Demonstrated practical applications in transportation and logistics.

Abstract

In this paper, we propose a paradigm for processing in parallel graph joins in road networks. The methodology we present can be used for distance join processing among the elements of two disjoint sets R,S of nodes from the road network, with R preceding S, and we are in search for the pairs of vertices (u,v), where u in R and v in S, such that dist(u,v) < {\theta}. Another variation of the problem would involve retrieving the k closest pairs (u,v) in the road network with u in R and v in S, such that dist(u,v) <= dist(w,y), where w,y do not belong in the result. We reckon that this is an extremely useful paradigm with many practical applications. A typical example of usage of our methods would be to find the pairs of restaurants and bars (in that order) from which to select for a night out, that either fall within walking distance for example, or just the k closest pairs, depending on the parameters. Another entirely different scenario would involve finding the points of two distinct trajectories that are within a certain distance predicate, or the k closest such points. For example, we would like to transfer from one train to another a few tones of freight, and hence, we want to minimize the distance we have to cover for moving the cargo from the carrying train to the other. We reckon that this endeavor of ours covers exactly those needs for processing such queries efficiently. Moreover, for the specific purposes of this paper, we also propose a novel heuristic graph partitioning scheme. It resembles a recursive bisection method, and is tailored to the requirements of the problem, targeting at establishing well separated partitions, so as to allow computations to be performed simultaneously and independently within each partition, unlike hitherto work that aims at minimizing either the number of edges among different partitions, or the number of nodes thereof.