Live Migration With Hardware Acceleration
summary
Offloads VM memory compression to a PCIe compression accelerator (Intel QAT) and plans to add DSA/DLA/IAA; implements zero copy DMA to guest memory, multi-page batching (up to 63 pages) and a cached request pool. Tests on Broadwell E5 show up to ~5x higher migration throughput vs CPU gzip-like compression, CPU compression uses ~678% CPU while QAT stays <40–70% CPU, and a random 128 GB VM that never finished with software compression completes in ~60 seconds with the accelerator. Future work adds VFIO zero copy and dynamic accelerator selection.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:21Agenda
- 0:54Project Goals
- 3:55The Migration Setup
- 8:37Multi-Page Process
- 11:23Multi-Page Processing
- 15:14Test Results
- 18:18Migration Throughput
- 24:29Smart Acceleration Support
related talks
17mDSA-2LM: A CPU-Free Tiered Memory Architecture with Intel DSA
31mHardware Friendly Vhost vDPA: Towards an Efficient and Migratable Device Model
41mOverview of Linux Page Migration and Accelerating It With Multi-threading and DMA Of...
31mDebugging Tools and Techniques for Virtualized/Containerized Automotive Systems
45mDynamic vCPU priority boosting in KVM for latency sensitive workloads
26mLessons Learned: Optimizing KVM Performance for EHR Systems
36mShared Virtual Addressing for High Performance Arm Infrastructure Platforms
19mSynQuacer use case update