Hyperconverged CEPH: Prioritize CPU over VMs for OSDs?

[michael.schaefers]

We have a 5-Host hyperconverged Proxmox 7.1 with CEPH as VM storage (5 SSD OSDs per host). My understanding is that CEPH I/O highly depends on available CPU power

Would it make sense to prioritize the OSD processes over any VM process to have (near-)full CPU power available for CEPH even under heavy CPU load?

If it makes sense, how would you do it in a clean and permanent way?
tuning the OSD cgroups (cpu.weight)?
tuning the nice level of OSD processes?
reserve CPU cores exclusively for OSDs?

 

[Tmanok]

Hi Michael,

I've also been interested in improving my available CEPH performance. Do the Proxmox engineers have any guidance here?

There are many performance tuning guides but I am not certain for their compatibility, for instance would modifying the following help?

osd_op_num_threads_per_shard = 1
TO
osd_op_num_threads_per_shard = 4

According to SUSE:

Increasing the number of op threads may be helpful with SSD and NVMe devices, as it provides more work queues for operations.

Thanks, looking forward to reading more experienced replies.


Tmanok

 

[Lephisto]

Now - has anybody tried this?