uDiscovery: An Urban-Centric Model for Service Discovery in Smart Cities
Citation:CABRERA, CHRISTIAN HUMBERTO, uDiscovery: An Urban-Centric Model for Service Discovery in Smart Cities, Trinity College Dublin.School of Computer Science & Statistics, 2020
uDiscovery-twoside.pdf (PDF) 98.69Mb
Cities offer services to their citizens to improve their overall quality of life (e.g., education, or health care services) and these services are frequently supported by digital information (e.g., library opening hours or hospital status) on the Internet. IoT devices have potential to further enhance city services as they can provide additional digital information from direct interaction with local urban spaces. These devices enable the collection of city and citizens' context, which may be useful to infer citizens' needs. For example, a citizen at a bus stop is more likely to need information about transportation city services. Service-oriented architectures (SOAs) are commonly-used to manage the services provided by IoT devices and include processes to register, discover, compose, execute, and monitor services. This thesis focuses on the discovery process, which is challenging for IoT services because they are fundamentally different from web services in the traditional Internet. In particular, the expected number of services is much greater, which will impact discovery efficiency, and they work in dynamic environments where distributed and adaptive architectures are more appropriate. Existing research on IoT and service-oriented computing (SOC) improves discovery efficiency by reducing search spaces, which is achieved by organising services into different groupings. Different approaches organise services according to different service attributes, such as location, or domain, which are captured in structures like overlays or hierarchies. However, approaches that organise services based on their attributes still create large search spaces where there are a large number of services, which is the case in smart cities. Moreover, the supporting structures do not provide enough information to drive the discovery process. For example, in a location-based approach, a request r1 might not get a response because service s1, which is relevant to r1, is in a different geographic area. In addition, overlays or hierarchical structures are static, but cities are dynamic and require continuous adaptation of their information systems. There are adaptive service discovery approaches that react to changes in service properties or the network topology, but they do not consider changes in the real-world environments with which services interact. Finally, current approaches to service composition are limited when constituent services need to be discovered from large IoT environments. Conversation-based approaches to service composition have good discovery accuracy but require high human intervention to define composition plans in advance. Interface-based approaches avoid human intervention but use expensive processes that affect discovery latency and accuracy. This thesis introduces uDiscovery, a distributed urban-centric model to support adaptive service discovery, designed to be efficient and adaptable in smart city environments. uDiscovery organises service descriptions based on urban-context. Gateways in a city environment recognise their surrounding places and create search spaces only relevant for these places. This urban context also drives service discovery by forwarding requests to gateways where they are most likely to be solved. uDiscovery adapts the service organisation as the city evolves. Each gateway recognises different city events as they occur and reacts to them by moving services from other gateways. This self-adaptive organisation puts the right service at the right place at the right time, in preparation for discovery. uDiscovery also includes a planner that searches for services that constitute compositions. The planner uses consumers' feedback to drive a progressive search that improves both discovery latency and accuracy. uDiscovery has been evaluated using a city simulation and an IoT test bed. Evaluation metrics include the discovery success rate, discovery accuracy, discovery response time, and the network overhead under varying number of services, and different mobility scenarios. Results present both the strengths and limitations of the proposed service discovery model. In general, uDiscovery outperforms baselines as it solves more requests with good accuracy and latency, at the cost of higher network overhead.
Science Foundation Ireland (SFI)
Publisher:Trinity College Dublin. School of Computer Science & Statistics. Discipline of Computer Science
Type of material:Thesis
Availability:Full text available