CISE Seminar: Yigong Hu, Boston University

Talk Title: Making Large-Scale Systems Self-Aware: AI-Assisted Performance Reasoning for Datacenter Applications

Modern datacenter applications operate under high concurrency, dynamic workloads, and increasingly complex internal logic. Ensuring stable end-to-end performance in such systems is fundamentally challenging because performance behavior is shaped by subtle interactions inside applications, including internal scheduling, buffering, and resource allocation. These internal mechanisms are largely invisible to the operating system, yet they critically determine application performance.

In this talk, I present an AI-assisted program analysis framework that exposes internal application signals and leverages them to reason about performance behavior. By combining large language models with static and dynamic program analysis, our approach identifies application-defined resources and tracks how requests interact with them. I will introduce two systems built on this idea. First, Atropos mitigates resource contention at runtime by identifying and controlling requests that drive overload. Second, GigiProfiler detects complex performance issues arising from misuse of internal application resources. Together, these techniques demonstrate how AI-assisted analysis can enable more intelligent, self-aware performance management in large-scale systems.

Yigong Hu is an Assistant Professor in the Department of Electrical and Computer Engineering at Boston University. He was previously a postdoctoral researcher at the University of Washington, where he worked with Prof. Baris Kasikci. He received his Ph.D. in Computer Science from Johns Hopkins University, where he was advised by Prof. Ryan Huang.

His research focuses on computer systems, system reliability, and machine learning systems. In particular, he develops principled system techniques to detect, diagnose, and mitigate performance issues in cloud systems, machine learning systems, and operating systems. His work has appeared in top systems conferences and has received awards, including the Best Paper Award at ASPLOS 2019.

Faculty Host: Ayse Coskun

Student Host: Jiatong Guo