IoT and Cloud Service Centric Framework for Enablement of Smart Cities

Abstract

Smart City is a new governance model that aims to efficiently manage the civic infrastructure and public services to enhance the city administration for overall benefit of citizens. Smart cities use inter-disciplinary and complementary technologies, such as Internet of Things (IoT), Big Data, Cloud Computing, and Edge Computing to address urbanization challenges like accessible transportation systems, efficient energy supply, insights-based urban planning, and so on. The synergistic integration of these technologies enable IoT-based data-driven intelligent smart city applications like smart traffic systems, smart street lighting, smart governance, and many more to achieve sustainable urban growth. To enable IoT-based data-driven smart city applications, a comprehensive study of big data management across several IoT applications is undertaken. The analysis of existing literature led to two outcomes – a taxonomy of big data management in IoT and identification and definition of 13 V’s challenges for big data in IoT. Further, current status of big data management in smart city frameworks is analyzed. The review of current works reveal that the state-of-the-art research works analyze and manage only the IoT big data, while geospatial data of a city is largely ignored. Since, multi-resolution geospatial datasets of a city represent spatial regions at different managerial levels, hence, using the geospatial data can lead to significant insights at spatial hierarchies for smart governance of the city. Therefore, this research undertakes to propose a cloud-based smart city framework named as, Cloud4IoTCity that considers the integration of city’s geospatial datasets with IoT big data for smart governance. The proposed Cloud4IoTCity framework successfully analyzes the IoT big data, which is available as batch data to enable smart governance applications in a request-response model. This proposed framework is enhanced to accommodate the real-time smart city IoT applications that operate in continuous sense-process-actuate control loop by utilizing edge computing layer. In this regard, the optimal placement of application services among the three-tier IoT-edge-cloud architecture to minimize latency and satisfy application’s diverse resource requirements is a major challenge, which is addressed in this work. In this regard, a Deep Reinforcement Learning (DRL) model named as, UrbanEnQoSPlace is proposed that solves the optimal service placement problem to minimize the overall latency and energy consumption for a set of smart city applications, while satisfying application’s diverse resource requirements in a smart city scenario. Cloud4IoTCity, a cloud-based framework that performs holistic big data management (i.e., store, process, analyze, visualize) and spatio-temporal analysis of IoT big data is proposed. The Cloud4IoTCity is designed to consider the fusion of IoT and geospatial data of a smart city and enable spatio-temporal analysis (both, descriptive and predictive) of fused urban data. Cloud4IoTCity comprises of four layers: storage layer, processing layer, service layer, and application layer. At first, the fusion of IoT big data with city’s geospatial data is done to build the storage layer of the framework on the cloud object store in the form of Delta Lake tables that are based on the recent Lakehouse storage architecture. Then, processing layer comprises of a customized compute cluster for distributed big spatial data processing of underlying fused data to support upper layer services. The service layer offers various data engineering and data processing services to the upper layer application. The data engineering service include, data fusion to create storage layer, add new data, z-order data layout optimization, time travel to a dataset version. Data processing services include spatio-temporal descriptive and predictive analysis, and map-based visualization services. Finally, the application layer features a cloud-based Software-as-a-Service (SaaS) application that enables end-users i.e., smart city managers to run service layer services on the underlying data. Further, to enable descriptive analysis in the form of spatio-temporal queries, a taxonomy of fifteen spatio-temporal queries is proposed, which defines five spatial, four temporal, and six spatio-temporal queries to understand urban dynamics from space-time dimensions. Secondly, to validate the proposed framework, a case study on urban traffic analysis for Dublin city is implemented that uses real IoT traffic dataset from Dublin. This IoT dataset comprises of three years (2020-2023) hourly traffic readings from inductive-loop sensors deployed at 500+ sites across the city. This IoT dataset is integrated with Dublin’s three geospatial datasets – Dublin postal districts (small areas), Dublin administrative areas (large admin zones) and Dublin city (entire city). Integration of IoT dataset with geospatial datasets of the city enables the spatio-temporal analysis at multiple scales of space-time resolutions. All the service layer services are implemented as PySpark scripts on a customized compute cluster having Delta Lake, Apache Spark and Apache Sedona software. The service layer services are accessible to end-users through the developed cloud-based web application named as Urban Analytics. The framework is validated for multi-resolution spatio-temporal descriptive analysis (query-based retrieval for proposed fifteen spatio-temporal query types) and predictive analysis (site-level hourly or daily traffic prediction). Experimental results show that distributed processing of queries reduces the average query runtime by 15.25% on a cluster with 4 worker nodes and by 28.12% on a cluster with 8 worker nodes as compared to a cluster with 1 worker node. Experimental results show that employing the z-order data layout optimization on data reduces the average query runtime by 34.12% as compared to the non z-ordered layout of data. Finally, the second aspect for smart city enablement i.e., edge-based efficient service delivery for real-time IoT applications in a smart city is addressed. Among the various edge computing paradigms, the Multi-access Edge Computing (MEC) has been considered in this research as the telecommunication networks have matured over the years and are seen widely deployed across every city. Further, this research considers federation of multiple MEC providers in a city to enable service placement seamlessly across multiple MEC vendors. To solve the service placement problem for a set of real-time smart city IoT applications in ‘Urban IoT - Federated MEC - Cloud’ architecture, UrbanEnQoSPlace DRL model is proposed. The proposed model is derived from the Dueling Deep-Q-Network to solve the placement problem by minimizing the overall latency and energy consumption for all applications, while satisfying their IoT requirements (sensors/actuators type and invoke-rate), resource requirement (cpu, memory) and per-flow latency, bandwidth requirements. Further, to enable the satisfaction of IoT locality constraints (i.e., match sensors, actuators capabilities), a novel model policy named as, ϵ-greedy with mask is proposed. The proposed policy adds a boolean mask on the standard ϵ-greedy policy, such that the valid placement nodes (model actions) for a sensing/actuating service are set to true, while others are set to false. This policy satisfies the IoT constraints by masking invalid placement nodes, which enables UrbanEnQoSPlace to converge faster. Experiments are conducted to validate the proposed UrbanEnQoSPlace DRL model against state-ofthe-art DRL algorithms. Experimental results for convergence analysis, reward analysis show the superior performance of UrbanEnQoSPlace. Further, UrbanEnQoSPlace is also validated for scalability by varying i) no. of applications ii) no. of application services and no. of placement nodes. Finally, experiments show that the proposed ϵ-greedy with mask policy results in 96.9% reduction in number of constraints violation as compared to the standard ϵ-greedy policy for solving the placement problem using UrbanEnQoSPlace model.