Urban cities faced with overpopulation are embracing data-intensive applications in order to maximize constrained resources. They utilize Internet-of-Things (IoT) devices deployed throughout the city, gathering data to be processed and analyzed in the cloud. However, the need for real-time services require a shift from the cloud towards edge and fog computing paradigms. In addition, growing concerns for privacy, trust and autonomy require moving away from centralized approaches to a more decentralized one.This edge-centric computing delegates processing to edge-devices.
A new framework for gathering, distributing, processing and aggregating tasks and results over heterogeneous devices must be created to achieve edge-centric computing. In this presentation, we discuss three challenges to realizing such a framework: 1) how to implement it over distributed devices, 2) how to ensure service resiliency, and 3) how to deploy smart city services on this framework.
For the first challenge, we describe our middleware based on this new framework. We show how it is deployed over distributed nodes by implementing a workspace recognition service. We introduce workflows for data distribution and aggregation, task allocation and decentralized execution. We show that in-situ resource provisioning on distributed nodes decreases execution time by 20% for every node added.
For the second challenge, we improve on the middleware and create a testbed that verifies the effects of anomaly detection and node configuration on service response times. We show that data falsification attacks can be prevented without additional burden on the system. We also show that varying distributed node configuration affects overall execution time.
For the third challenge, we realize a complete smart city service on the middleware. We develop a distributed route planning service with a task allocation algorithm that utilizes city road side units as distributed computing nodes. We explore the feasibility of the service through simulation and show that neighbor grids have a positive effect on processing time. We emulate this with real-world data, and show that our routing system with task allocation algorithm is able to decrease processing time by 50% with only a 7% decrease in travel time accuracy.