Grant: Discovery Grants Program - Individual

This research program provides mechanisms and protocols to enable the automatic elasticity of stateful data stream processing applications on edge computing infrastructure. Several application scenarios nowadays, including autonomous vehicles, Industry 4.0, and the Internet of Things, produce streams of data events that require timely processing. Handling state and processing data at the Internet edge can minimize the end-to-end latency of handling events, which is particularly useful for emerging telemetry applications that create digital twins of complex equipment or entire user communities. While the elasticity of data stream processing has been examined in the academic literature, the existing work either focuses on applications with stateless operators or does not consider the heterogeneity and dynamicity commonly present in edge environments. The elasticity of stateless data stream processing applications is often more straightforward than stateful ones. Still, it seldom represents real-world data stream processing where operators maintain state (e.g., windows, joins, and machine learning models). We choose to focus on edge computing because we believe that addressing the data stream processing under such challenging conditions can enable the deployment of data stream processing along a large infrastructure spectrum. The long-term goal of this research program is to enable a unified framework that allows data stream processing along the entire cloud-edge-device continuum. We will build on our previous work on data stream processing on the cloud but focus on the edge component of the cloud-edge-device continuum at this stage. The proposed methodology will: 1. Create techniques to express, checkpoint, and migrate state in edge computing settings while accounting for resource heterogeneity and network and energy constraints. 2. Conceive repartitioning methods for large key spaces that guarantee in-order event processing and consider the memory and energy limitations of edge resources. 3. Build a lightweight application monitoring and profiling framework that identifies performance issues. 4. Create mechanisms and protocols to enforce responsive scale-out/scale-in decisions at edge computing infrastructure. 5. Design AI policies for making automated elasticity decisions. The program will contribute to science and engineering knowledge by introducing novel methods and a framework for handling the elasticity of stateful data stream processing at the Internet edge. The framework will allow more efficient use of edge computing infrastructure and enable a new range of services to end users. Canadian and international corporations can use the outcomes of this project to design better service offerings on edge computing and data stream processing. The project will also help train the next generation of HQP and advance in areas that are key to industry leaders such as Amazon, Microsoft, Google, and IBM.