-
Story
-
Resolution: Done-Errata
-
Normal
-
rhel-9.3.0
-
libvirt-10.0.0-1.el9
-
Medium
-
rhel-sst-virtualization-storage
-
ssg_virtualization
-
20
-
26
-
3
-
QE ack, Dev ack
-
False
-
-
Yes
-
Red Hat Enterprise Linux
-
None
-
Pass
-
Automated
-
Enhancement
-
-
Done
-
-
Unspecified
-
10.0.0
-
None
Work is underway to add per-virtqueue IOThread mapping to QEMU's virtio-blk and virtio-scsi devices.
Currently each device instance can be associated with just one IOThread where all virtqueues are processed. This is known to be a bottleneck and prevents true multi-queue operation because all I/O requests are funneled through a single IOThread on the host.
The new multi-queue support in QEMU will allow individual virtqueues to be mapped to an IOThread. It is likely that most users will wish to spread virtqueues round-robin across a set of IOThreads.
New domain XML syntax is necessary to expose this feature once it becomes available in QEMU.
I am filing this BZ now so there is time to discuss the feature and so that libvirt support can be developed in parallel to QEMU.
- depends on
-
RHEL-20764 [qemu-kvm] Enable qemu multiqueue block layer support
- Closed
- is blocked by
-
RHEL-15267 Rebase libvirt in RHEL-9.4.0
- Closed
- is related to
-
CNV-38292 Enable Qemu multiqueue block layer
- Stakeholder Acceptance
- relates to
-
RHEL-23607 Update device with iothread mapping changed returned success but no change happened
- Release Pending
- external trackers
- links to
-
RHBA-2023:125049 libvirt bug fix and enhancement update
- mentioned on