MigrOS: Transparent Live-Migration Support for Containerised RDMA Applications
summary
MigrOS provides transparent live migration for containerised RDMA applications by making minimal changes to the RoCE v2 protocol and kernel RDMA stack so user binaries require no modification, preserving no-overhead during normal operation. Prototype integrates with CRIU, ib_core and device drivers and implements adds two new RDMA states plus resume/stop messages, tested on SoftRoCE (kernel module) and shows migrations of MPI jobs in 0.68–1.2 s and up to 33% lower latency than software-only interception.
AI-written summary · about
chapters (10)
- 0:00Intro
- 2:08Live Migration
- 2:46RDMA Networks
- 4:32Consistent Checkpointing
- 5:27What is a Queue Pair (QP)?
- 6:15Pause/Resume Protocol
- 9:23Implementation
- 11:54Evaluation: Motivation
- 13:11Evaluation: Practicality
- 14:24Conclusion
related talks
27mImplementing SR-IOV Failover for Windows Guests During Migration
39mContainer Live Migration
20mCRIU and SELinux
27mLive Migration With Hardware Acceleration
37mCRI-O: Under the Hood
44mEasy Geo-Redundancy with MARS
19mRPMsg to Accelerate Transition Between Multi-SoC and Multi-processor SoC Solutions
13mRosenBridge: A Framework for Enabling Express I/O Paths Across the Virtualization...