1. System Performance Modeling

We develop and release innovative methodologies for analyzing and modeling computer systems. These methodologies allow us to understand the performance of various commodity systems, including processors, storage systems, and datacenters, even when their internal mechanisms are not fully disclosed.

2. ML for System (System Autotuning)

Modern systems like databases and storage are becoming increasingly complex. They often have a large number of internal tuning knobs that can be configured in countless ways. In this research area, we leverage machine learning to automatically find the optimal configuration for a given system, leading to improved performance.

3. AI System Infrastructure

The ever-growing size and complexity of large-scale AI models, such as large language models (LLMs), pose significant challenges for existing system infrastructure. Our research aims to design new system infrastructures specifically optimized for handling large-scale AI workloads.