Speeding Up VM’s I/O Sharing Host's io_uring Queues With Guests
summary
Demonstrates a working proof‑of‑concept that shares host io_uring queues by mapping SQ/CQ mapped in guest memory and using an eventfd plus io_uring restriction lists to safely limit ops; measures show the passthrough is faster than virtio/vhost alternatives and approaches bare‑metal as concurrency increases (worst case <13% gap with one in‑flight I/O, gap goes to zero with many in flight). Missing pieces: guest memory translation for host io_uring and upstream patches; next steps include a VDPA block software device.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:23Agenda
- 1:54io_uring system calls
- 3:04Submission and Completion Queues
- 4:36Resources registration
- 5:51Polling
- 6:56QEMU and io_uring
- 7:28virtio block
- 9:44io_uring changes for passthrough
- 10:51io_uring restrictions
- 13:04io_uring passthrough PoC performance
- 14:43vhost block
- 17:00VDPA block software device
- 18:17Next Steps
related talks
40mAchieving the Ultimate Performance with KVM
26mAnalysis of AMD HW-assisted vIOMMU Implementation and Performance
20mPassthrough/Headless GPU Gets Ahead
35mIntroduce an SPDK vhost FUSE Target to Accelerate File Access in...
25mXen and XenServer Storage Performance
21mNVMe Emulation Performance Optimization
20mMulti-Queue Fair Queuing
38mScaling IO-bound application in the cloud