Speaker
Description
Introduction
Smart city is a philosophy focused on improving many aspects of life in a modern city while consuming less energy. It includes optimization of traffic, efficient control of energy consumption, easier access to public services and similar topics.
The poster is focused on the traffic optimization aspect of the smart city by providing an easy-to-use service based on an experimental self-adaptive navigation system. This system combines historical and current monitoring data using methods based on the graph theory. In our approach, we use a custom implementation of betweenness centrality score. Its application on a graph representing a large number of road segments is computationaly demanding. Therefore, we use a high-performance computing (HPC) infrastructure to provide the most current data as fast as possible. Efficient operation of the system is maintained by technologies developed within the ANTAREX project, mainly by an auto-tuning mARGOt framework, and LARA DSL language together with its compilation infrastructure.
What-If Analysis for Smart City
Cities need to know what would happen when a road is closed, or lane reduced due to a road work or a major incident. Our solution enables nearly real-time estimation of the traffic network state which can be affected by various traffic events. This information can be later used to implement intelligent navigation policies to mitigate the traffic congestion risks and optimize the traffic flow.
We extended the original betweenness centrality algorithm by the concept of the vertex importance. It allows us to assign an importance value to the vertex, thus better approximate its behaviour in the network and to make the adjustments in response to a current traffic situation. The extension is based on two parameters (source and destination vertex importance) representing situations, that can occur in the network. Modification of the values leads to the change of the betweenness score on the shortest paths leading away from/ or towards the vertex. In a practical sense, it is used to identify a place where traffic originates, like a residential area in the morning, or a location that has or is about to have an increased incoming traffic, like a sport facility before the start of the important event.
We have tested different values of the both parameters and their influence on the traffic network. We have also tested if the modification affects the execution time on compute nodes. The tests were performed on a map with 225,526 vertices and 529,767 edges. The modification showed on average 8.76 % increase in runtime when compared with unmodified version. Memory usage was on average 4.04 % higher.
ANTAREX Tools for Traffic Modelling
LARA and its compiler Clava are used along the entire tool chain. All the possible optimizations for the specific betweenness centrality implementation (i.e. memoization, custom precision, and number of OpenMP threads) are introduced into the code as software knobs. At runtime, autotuning tool mARGOt can decide to include or not, and eventually to configure the SW knobs, according to profiling information, type of analysis to be performed, and dimension of the graph.
