Xudong Sun

The goal of my research is to make computer systems reliable. Even critical computer systems have bugs, and the bugs come with serious consequences, e.g., outage of your favorite app, loss of your personal data. My research is about making sure such bugs never happen (again)!

I enjoy building systems, hacking systems, testing systems and writing mathematical proof for systems. My research is in the intersection of computer systems, software testing and formal methods, and I take a hybrid approach to systems reliability — I build (1) automated and systematic testing techniques to detect bugs in widely-used systems and develop a deep understanding of these systems, and (2) practical formal verification techniques to build provably correct, production-grade systems based on insights learned from testing. My research involves techniques including deductive verification, model checking, fault injection, and testing.

As a system researcher, I care about real-world systems that support everyone's daily life, and developers who struggle with bugs and failures of these systems. My research has improved the reliability of many critical systems, including Kubernetes, ZooKeeper and Hadoop, by finding bugs in them. We have presented at industry venues including KubeCon and SRECon on some of the emerging reliability challenges developers have to face, and how our research can help.

My Ph.D. research focused on the reliability of cloud systems. In particular, I have built Anvil [OSDI'24], a framework that allows developers to build practical cluster managers (e.g., Kubernetes) and formally verify high-level properties (e.g., liveness and safety) of their implementations. The verified cluster managers can be deployed in real-world clusters and manage real-world applications and services. Anvil is inspired by my previous work Sieve [OSDI'22] and Acto [SOSP'23] on testing cluster managers. I have been continuously developing and improving Anvil to make it an open platform for practitioners and researchers to develop cluster managers and verify various liveness and safety properties.

In addition to cluster managers, I have also worked on other layers and aspects of modern cloud stack, including consensus-based systems (e.g., ZooKeeper, etcd), cloud service (e.g., Azure Storage, AWS S3) integrations, and cloud system configurations.

I believe that systems research should be relevant, practical and beautiful. That's what I always aim for.

Xudong Sun

Research

Recent news

Publications

Service

Teaching

Mentoring

Honors & Awards

How to pronounce my name