Containment Domains (CDs)

Summary

Containment Domains (CDs) were the core of my PhD research at UT Austin (advised by Mattan Erez). CDs are a programming construct and runtime system that let applications express hierarchical, application-tailored resilience as an alternative to traditional global checkpoint/restart.

I developed the CD runtime system, analytical model, and supporting tools, and enabled and evaluated CDs across MPI, CUDA, and Legion programs.

Highlights

• Designed and built the CD runtime system for MPI programs.
• Developed an analytical model and tooling to reason about resilience overhead.
• Extended the approach to GPU and task-based execution (see related projects).

Links

• Demo of CD Runtime for MPI Programs (video, 2016)
• CD API documentation
• Related publication: Containment Domains Semantics v0.2 (Technical Report, 2014)
• Legacy project page: kyushick.github.io/project/cds/