MigrOS - Transparent Live-Migration Support for Containerised RDMA Applications

Explore a conference talk from USENIX ATC '21 that introduces MigrOS, an OS-level architecture enabling transparent live migration of containerised RDMA applications. Delve into the challenges posed by RDMA networks bypassing the OS for improved latency and bandwidth, and discover how MigrOS overcomes these obstacles to facilitate live migration. Learn about the minimal changes to the RDMA communication protocol that allow for migration without compromising critical path operations. Examine the implementation details, including the concept of Queue Pairs (QPs) and the Pause/Resume Protocol. Evaluate the practical implications and performance benefits of MigrOS, including its ability to achieve up to 33% lower network latency compared to software-only techniques. Gain insights into consistent checkpointing and the broader impact of this innovation on containerized RDMA applications.