PhD defence by Yiwen Wang

Yiwen Wang PhD defence banner

Title

Scalable and Reactive Data Management for Mobile Internet-of-Things Applications with Actor-Oriented Databases

Summary

Development and utilization of Internet-of-Things (IoT) applications are increasing at an unprecedented speed in many domains, such as supply chain and logistics tracking, smart agriculture, e-health, and intelligent transportation systems, to name a few. Among these applications, mobile IoT applications have received special attention by virtue of having an ever more significant impact on people's day-to-day and society. Mobile IoT applications employ widely used portable and movable IoT entities that collect data, share information, and interact with each other to achieve common goals. While a host of data management solutions may be employed in the architecture of IoT applications, ranging from edge data processing to historical analytical database systems, a particularly interesting component is the IoT data platform, which is a cloud-resident infrastructure for online management of IoT data.

The growing connection coverage and increasing requirements in mobile IoT applications bring about special problems and challenges to be explored in IoT data platforms. Firstly, scalability is a necessary requirement for an IoT data platform because of the explosive growth of the IoT. Secondly, tight low-latency reactivity in an IoT data platform is required while managing a massive amount of highly concurrently generated data. Thirdly, support for heterogeneous data types is demanded in an IoT data platform due to the variety of IoT devices. Fourthly, elasticity is also required in an IoT data platform to handle dynamically changing IoT workloads. Last but not least, guidance on ensuring the dynamic and flexible development of IoT data platforms is crucial for application developers.

Existing approaches have explored how to support different properties required in IoT data platforms, but they fall short of fulfilling our goal of providing programmable, scalable, and reactive data management for mobile IoT applications. Recently, Actor-Oriented Databases (AODBs) have been proposed as a new cloud-based data management architecture for distributed, scalable, stateful, and interactive applications. We find that AODBs are naturally suitable for satisfying the requirements and solving the challenges of building IoT data platforms. Therefore, we inspect the distinct requirements of building IoT data platforms through two case studies, and we investigate how to effectively model and build IoT data platforms with AODBs. Then, we focus on easing the complexity of building mobile IoT data platforms with AODBs while providing scalability and reactivity. We propose a novel class of data management systems called Moving Actor-Oriented Databases (M-AODBs), which integrate the new abstraction of moving actors for reactive moving objects with two data concurrency semantics for different application use scenarios. A first implementation, Dolphin, is presented to validate the design of M-AODBs. Afterward, to handle the typically skewed spatial distributions in mobile IoT applications, we apply varied classic spatial partitioning techniques in Dolphin to evaluate, analyze, and manage bottlenecks due to data skew. Our experimental study illustrates the impact of spatial partitioning in Dolphin and unveils several promising directions for future work.

Assessment Committee

  • Associate Professor, Boris Düdder, Department of Computer Science, UCPH
  • Research Professor/Associate Professor, Simonas Šaltenis, Vilnius University (Lithuania)/Aalborg University (Denmark)
  • Senior Researcher, Fabio Porto, National Laboratory of Scientific Computing, LNCC (Brazil)

Academic Supervisor(s) / Moderator of this defence

Associate Professor, Marcos António Vaz Salles, SDPS Section, Department of Computer Science, UCPH

This defence will be conducted online.

For a digital copy of the thesis, please go to our PhD page