Research Summary

I work on computer architecture and systems, with a primary focus on efficient and resilient datacenter management optimization. My research asks how to model, observe, schedule, and harden emerging AI and cloud infrastructures under hardware heterogeneity, resource constraints, and frequent failures.

Research Themes

• High Energy Efficiency: modeling, allocating, and scheduling heterogeneous or constrained hardware resources to improve performance per watt.
  • Processor-scale CPU-based LLM serving: APPT’25a, HPCA’26
  • Device-scale autonomous embedded systems: ICCD’24, TC’26, TACO’24, RTSS’24
  • Datacenter-scale cloud-native workloads: SoCC’23, IPDPS’22, IPDPS’23, SCIS’25
• High Resilience: analyzing, detecting, and eliminating hardware and software failures to keep large-scale infrastructure stable.
  • Low-overhead observability for shared datacenters: ASPLOS’25, SCIS’26
  • Efficient hardware fault tolerance for AI infrastructure: ongoing
• Agentic Infrastructure Management: exploring agentic AI methods for CPU-centric infrastructure modeling, design-space exploration, and high-resilience management.