With regard to:
https://forum.proxmox.com/threads/ceph-read-performance.25785/page-3
I don't see on the roadmap parallelization of disk IO, but recognize that it would have a substantial benefit to small-medium sized ceph clusters, which tend to have low single threaded IO.
Currently, I've worked around the issue by attaching the same (scsi-single) disk multiple times to single machines. (4 nodes, 32 identical osds)
This works relatively well alongside dm-multipath to spread io load over multiple threads.
However, there appears to be a limit on total scsi disks that can be attached to a single machine (30), which means this strategy does not work in many cases. (Bug Report/Feature Request)
As well, the management of multipath, multipath boot devices, and cli vm configuration updates are each a complexity that most users wouldn't consider to be convenient.
Is there still plans to enable multi-threaded disk IO?
Is this even a proxmox issue, or is it further up?
Is there a better way to approach the problem currently?
Is there a proxmox recommended optimization path/documentation for the low single threaded IO issue?
https://forum.proxmox.com/threads/ceph-read-performance.25785/page-3
I don't see on the roadmap parallelization of disk IO, but recognize that it would have a substantial benefit to small-medium sized ceph clusters, which tend to have low single threaded IO.
Currently, I've worked around the issue by attaching the same (scsi-single) disk multiple times to single machines. (4 nodes, 32 identical osds)
This works relatively well alongside dm-multipath to spread io load over multiple threads.
However, there appears to be a limit on total scsi disks that can be attached to a single machine (30), which means this strategy does not work in many cases. (Bug Report/Feature Request)
As well, the management of multipath, multipath boot devices, and cli vm configuration updates are each a complexity that most users wouldn't consider to be convenient.
Is there still plans to enable multi-threaded disk IO?
Is this even a proxmox issue, or is it further up?
Is there a better way to approach the problem currently?
Is there a proxmox recommended optimization path/documentation for the low single threaded IO issue?