Abstract
As the real world continues to evolve dynamically, current and emerging IoT applications increasingly depend on data streams. These applications, typically deployed on resource-constrained edge devices, demand stringent requirements, particularly in terms of performance and resource budget. This thesis addresses these challenges by investigating stream compression techniques and systematically optimizing them on modern multicore edge hardware. Stream compression plays a vital role in reducing bandwidth usage and transmission overhead in IoT applications. However, the added compression overhead and the limited hardware resources of IoT devices necessitate a sophisticated software-hardware co-design strategy. Therefore, we propose CSTREAM, a software-hardware co-design stream compression framework tailored for IoT applications. By heuristically considering the computational behavior of stream compression and the properties of edge hardware during the parallelization of stream compression on multicore edge devices, CSTREAM achieves an optimal balance of high compression ratio, high throughput, low latency, and low energy consumption. Our comprehensive evaluations demonstrate the superiority of our optimization techniques and provide valuable insights for the future design of stream data management systems for AI.