Virtio-(balloon|pmem|mem): Managing Guest Memory
summary
Inflating a balloon from an 8 GiB VM to 4 GiB reduces migrated RAM to 4 GiB; migration and overcommit instead use free page hinting (host write-protects and reclaims large pages) and free page reporting (guest reports and waits for host processing). File-backed direct mapping via virtio-pmem moves the page cache to the hypervisor and forces fsync on flushes but only supports raw images and has persistence caveats. Fine-grained resize with virtio-mem uses megabyte-scale blocks (4 MiB on x86_64), is slower than ballooning and not fully production-ready.
AI-written summary · about
chapters (16)
- 0:00KVM FORUM
- 0:38Motivation: Manage Guest Memory
- 3:18Recap: Memory Ballooning
- 4:42Using Memory Balloon Inflation / Deflation ?
- 6:06virtio-balloon: inflate/deflate under guest control
- 8:07free page hinting: by Wei Wang (Intel)
- 9:10free page hinting: pros and cons
- 9:50free page reporting: by Alexander Duyck (Intel)
- 10:40free page reporting: pros and cons
- 11:19Balloon: TODOS
- 12:42virtio-pmem: Overview (2)
- 15:20virtio-prem: (Dis)Advantages and Open Items
- 18:31virtio-mem: Overview (1)
- 22:24virtio-mem: (Dis)Advantages and Open Items
- 25:35Summary + Outlook
- 29:01Resources
related talks
21mLessons Learned Building a Production Memory-Overcommit Solution
10mExtMem: Enabling Application-Aware Virtual Memory Management for Data-Intensive...
17mBreak On Through to the Other Side: Pooling Memory Elastically with RamRyder
52mChanging shared_buffers on the fly
30mLarge-Scale Practice of Persistent Memory in Alibaba Cloud
45mEnhancing PostgreSQL Plasticity: New Frontiers in Memory Management
26mLessons Learned: Optimizing KVM Performance for EHR Systems
35mDeveloping Software for Persistent Memory