Doctor of Philosophy (PhD)


Division of Computer Science and Engineering

Document Type



Modern data centers are the backbone of today’s Internet-based services and applications. With the explosive growth of the Internet data and a wider range of data-intensive applications being deployed, it is increasingly challenging for data centers to satisfy the ever-increasing demand for high-quality data services. To relieve the heavy burden on data center systems and accelerate data processing, a popular cost-efficient solution is to deploy high-speed, large-capacity flash-based cache systems. However, we are facing multiple critical challenges from device hardware, systems, to application workloads. In this dissertation, we focus on designing highly efficient caching solutions to cope with the explosive growth of data-intensive workloads and the huge traffic of data queries in data center systems. To realize high caching efficiency, we have addressed three critical issues, from data reliability, system performance, to cost effectiveness, mainly from the perspective of system design. Benefiting from the expressive object-based interface, we first present a highly reliable, efficient, object-based flash cache system, called Reo, to address the reliability issue in flash caching in data centers. To accelerate query processing on emerging time-series data workloads, we present a high-speed, large-capacity flash based cache system, called TSCache, by exploiting the unique properties of time-series data workloads. Finally, we further present an efficient hardware-accelerated compression scheme, called Taco, to significantly expand the usable cache space thereby improving the cost-efficiency and caching performance. Our experimental results demonstrate that our flash-based cache solutions can significantly improve the system performance, reliability, and cost efficiency. We hope our study can inspire the community to effectively address the emerging challenges and fully exploit the new opportunities in future data center systems.

Committee Chair

Chen, Feng

Available for download on Tuesday, November 01, 2022