As computing environments grow increasingly heterogeneous, ensuring seamless application execution across different CPU architectures has become a critical challenge. This talk introduces Dapper, a lightweight system that enables transparent process checkpointing and restoration across heterogeneous architectures.
Dapper leverages Linux’s CRIU mechanism to checkpoint a live process, rewrites its architecture-specific execution state, and restores the program on a different CPU architecture. By transforming the process externally and requiring only minimal compile-time metadata, Dapper avoids intrusive code modifications. This approach makes Dapper a versatile solution for various scenarios, including live process migration to improve energy efficiency on heterogeneous processors.
In this talk, Dr.Wang will present Dapper’s design, highlight key technical challenges, and demonstrate its effectiveness through real-world applications involving server workloads and benchmark suites. A demo of Dapper is available at: https://dapper-project.github.io/
Xiaoguang Wang is an Assistant Professor in the Department of Computer Science at the University of Illinois Chicago (UIC). His research focuses on computer systems and software security, with an emphasis on systems support for emerging computer architectures and secure software systems design. Recently, he has been exploring innovative ways to address systems and security challenges using large language models (LLMs).
Dr. Wang is the lead developer of Dapper, a framework that enables transparent process checkpointing and restoration across heterogeneous CPU architectures. Dapper addresses critical challenges in cross-architecture process migration by combining runtime state lifting, state translation, and architecture-aware memory management. His work enhances system resilience, improves fault tolerance, and facilitates workload migration across diverse hardware platforms.
Click here for Meetup registration
Additionally, complete Zoom link for online meeting
