Hello.
We are testing our RAID engine with a PVE instance. As our main goal is a performance, we sometimes have bottlenecks within system components that are not related to the storage device itself. E. g., CPU core may be the slowest component in the datapath.
QEMU 9 introdused the new feature that allows to utilise multiple CPU cores within one virtual storage devices.
There are 3rd-party implementations that allow the new feature to be used within PVE. We can confirm that iothread-vq-mapping make things faster. The caveats are the following:
Thank you.
We are testing our RAID engine with a PVE instance. As our main goal is a performance, we sometimes have bottlenecks within system components that are not related to the storage device itself. E. g., CPU core may be the slowest component in the datapath.
QEMU 9 introdused the new feature that allows to utilise multiple CPU cores within one virtual storage devices.
There are 3rd-party implementations that allow the new feature to be used within PVE. We can confirm that iothread-vq-mapping make things faster. The caveats are the following:
- The implementation mentioned is not a part of PVE product, therefore any PVE update may require reapplying the patch with potential incompatibilities.
- iothread-vq-mapping setting up requires manual editing of a VM configuration file, this is not a straight forward action that may lead to typos or other human element issues.
Thank you.
