Enhancing Timeliness and Power-Efficiency of Real-Time Data Services

Primary Investigator

• KD Kang, Professor, Department of Computer Science, State University of New York at Binghamton

Project Summary

Real-time data services play a crucial role in supporting data-intensive real-time applications, such as smart transportation, healthcare, and manufacturing, with significant societal value. These applications require cost-efficient real-time embedded databases capable of processing real-time data service requests in a timely fashion, using fresh data that represent the current real-world status, while also conserving power. This project aims to investigate novel methodologies to promote the progress of science in this essential yet underexplored area of research. Furthermore, by improving resource and power efficiency, it aims to considerably alleviate the challenges of deploying real-time data services in deeply embedded systems that have limited resources. The broader impacts of the project include training undergraduates and graduates, including underrepresented groups of students, outreach activities, and open-source software.

The goal of this research project is to bridge the knowledge gap on power-efficient real-time data services by investigating data-centric approaches that can improve both timeliness and power efficiency without compromising either. Achieving this objective, however, is challenging due to several factors, including varying arrival rates and data needs of user transactions based on the current real-world status, data and resource contention, and stringent timing, resource, and power constraints in embedded systems. To tackle such challenges, the research will (1) investigate a new effective real-time transaction model and data-centric approaches to reducing workloads that identify actual data requirements and freshness needs to enhance timeliness and power efficiency; (2) investigate a self-adaptive control framework that can maintain desired timeliness despite workload variations, using fewer processor cores to conserve power; and (3) explore how to leverage advanced memory hardware features and real-time data characteristics to further enhance timeliness and reduce processor and memory power consumption.

Publications

• Rahim Hossain, Tawheedul Islam Bhuian, Kyoung-Don Kang, TDA-L: Reducing Latency and Memory Consumption of Test-Time Adaptation for Real-Time Intelligent Sensing, Sensors, 2025, 25(12), 3574. [paper] [source code]

• Y. Liu, A. Andhare, K. D. Kang, Corun: Concurrent Inference and Continuous Training at the Edge for Cost-Efficient AI-Based Mobile Image Sensing, Sensors, 2024, 24(16), 5262. [paper] [source code]

• Y. Liu, K. D. Kang, AROD: Adaptive Real-Time Object Detection Based on Pixel Motion Speed, IEEE 100th Vehicular Technology Conference (VTC2024-Fall), Washington DC, USA, October 7-10, 2024. [pdf] [code]

• K. D. Kang, Limitations of Validity Intervals in Data Freshness Management, arXiv preprint arXiv:2407.20431, 2024.

• Y. Liu, K. D. Kang, Filtering Empty Video Frames for Efficient Real-Time Object Detection, Sensors, 24(10):3025, 2024. [pdf] [code]