Graduation Semester and Year
Summer 2025
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Computer Science
Department
Computer Science and Engineering
First Advisor
Hong Jiang
Second Advisor
Hao Che
Third Advisor
Song Jiang
Fourth Advisor
Upendranatha Sharma Chakravarthy
Abstract
Server applications operating in oversubscribed cloud environments face the dual challenges of meeting strict Quality-of-Service (QoS) requirements and improving resource and energy efficiency. As the number of user connections and workload diversity continue to grow, existing scheduling mechanisms struggle to balance QoS guarantees, fairness, resource efficiency, and power consumption. This dissertation presents a unified, cross-layer framework to address these challenges through three key contributions: AppleS, UTSLO, and REEF.
First, we propose AppleS, a user-space QoS-aware fine-grained I/O scheduling framework that delivers fair and efficient service to concurrent client connections. AppleS introduces a QoS-centric metric that guides admission control and scheduling decisions, adapting dynamically to changing workload characteristics. It achieves near-optimal throughput-latency tradeoffs with low overhead and is compatible with existing server architectures such as MySQL, MongoDB, and PostgreSQL. Experimental results demonstrate that AppleS improves tail latency, I/O fairness and overall throughput under high connection counts without compromising user experience.
Second, we develop UTSLO, a tenant-targeted latency-aware SLO enforcement framework that accommodates differentiated QoS requirements under multi-tenant environments. Unlike traditional percentile-latency-based resource allocation schemes, UTSLO enforces explicit tenant-customized latency-distribution-sensitive user disengagement ratio (UDR) targets through a closed-loop control mechanism, which adapts to workload shifts and efficiently ensures predictable QoS under multi-tenant workloads in real time. UTSLO maximizes overall throughput for best-effort users while strictly meeting the UDR targets for UDR-sensitive tenants, even under heavy oversubscription.
Finally, we introduce REEF, a runtime framework for Resource-Efficient Energy-aware Fair scheduling, that bridges connection-level scheduling with thread-level CPU and power management. REEF redefines the thread execution model via signal-driven Ideal Batch activation, converting self-serving worker threads into on-call threads. This enables predictable idle-time exposure for deep CPU C-state residency (e.g., C6), reducing energy consumption while isolating connection-specific QoS enforcement from power management schemes. REEF supports plug-in integration with AppleS, UTSLO, and other scheduling policies, and introduces a QoS-centric workload consolidation mechanism to isolate latency-sensitive workloads at the core level.
Across extensive evaluations using real-world workloads or benchmarks (e.g., YCSB on MongoDB and OLTP on MySQL), our proposed techniques collectively achieve up to 5.33 times energy-efficiency improvements, reduce tail latencies by up to 83%, and cut CPU resource consumption by up to 73.65%, all while maintaining or improving total throughput and strict SLO compliance.
This dissertation advances the state of the art in resource- and energy-efficient, QoS-aware scheduling by delivering modular, interoperable mechanisms that span network I/O, application, and OS resource management layers, addressing critical needs in modern cloud server platforms.
Keywords
QoS-Aware Scheduling, Resource and Energy Efficiency, Service-Level Objective (SLO), User Disengagement Ratio (UDR), Dynamic Power Management (DPM)
Disciplines
Computer and Systems Architecture | Data Storage Systems | Power and Energy
License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.
Recommended Citation
Li, Ning, "AN SLO-AWARE, MULTI-PRONGED APPROACH TO ENHANCING RESOURCE AND ENERGY EFFICIENCY IN SERVER APPLICATIONS" (2025). Computer Science and Engineering Dissertations. 421.
https://mavmatrix.uta.edu/cse_dissertations/421
Included in
Computer and Systems Architecture Commons, Data Storage Systems Commons, Power and Energy Commons
Comments
I would like to express my sincere gratitude to my supervising professor, Dr. Hong Jiang, for his continuous encouragement, insightful guidance, and unwavering support throughout the course of my doctoral studies. I am also deeply thankful to my academic advisors, Dr. Hao Che, Dr. Song Jiang, and Dr. Upendranatha Sharma Chakravarthy, for their valuable feedback, interest in my work, and for generously serving on my dissertation committee.