Title: Customizing System Software for Performance, Flexibility, and Isolation

Date: Tuesday, April 22, 2025

Time: 11:00 AM – 12:30 PM EST 

Location (Hybrid): CODA 0915 and MS Teams (link)

Sujin Park

Ph.D. Student

School of Computer Science

College of Computing 

Georgia Institute of Technology 

Committee: 

• Dr. Taesoo Kim (advisor) – School of Cybersecurity and Privacy, Georgia Institute of Technology
• Dr. Anand Iyer - School of Computer Science, Georgia Institute of Technology
• Dr. Ada Gavrilovska - School of Computer Science, Georgia Institute of Technology
• Dr. Weiteng Chen – Microsoft Research

Abstract: 

Modern computing systems face increasingly diverse workloads, heterogeneous hardware platforms, and stringent isolation requirements. Traditional operating system designs, optimized primarily for generality, often fall short in addressing scenarios with specific workload demands, performance goals, hardware capabilities, or security constraints. This thesis explores systematic approaches for customizing system software components to better meet these diverse, and sometimes conflicting, design goals.

First, flexibility and performance at the software level are addressed by bridging the semantic gap between applications and kernel behaviors. SynCord, a framework for application-informed kernel synchronization primitives, enables developers to dynamically customize kernel locks from user-space. By facilitating fine-grained, workload-specific kernel locks, SynCord significantly enhances performance and fairness in scenarios where traditional kernel locks fall short. Second, recognizing the critical role of hardware-aware customization, this thesis explores the design of secure system software that fully exploits emerging hardware features provided by the open RISC-V architecture. The RISC-V WorldGuard project demonstrates a flexible and scalable Trusted Execution Environment, dynamically adapting to stringent isolation requirements in sensitive applications such as robotics. Finally, building upon these targeted software- and hardware- specific customizations, the thesis proposes a general methodology for system performance optimization. We formalize foundational methodologies—batching, caching, reordering, and specialization—providing a comprehensive basis for optimizing sequential system performance. Additionally, SysGPT, a GPT-driven assistant trained on these optimization principles, is introduced to automate and guide developers in applying these strategies effectively across various contexts.

Collectively, these contributions illustrate a cohesive framework for customizing system components across multiple layers, guided by specific application goals, hardware features, and systematic performance optimization principles. The thesis demonstrates substantial improvements in flexibility, performance, fairness, and security, paving the way for future adaptive and specialized operating systems.