1 ceph or 2 ceph clusters for fast and slow pools?

[chell]

Hi All

Is there any recommendations for configuring pools with differing goals?

I am setting two pools with opposite goals. A fast NVMe pool for VM's and a slow cephfs using spinning drives. Considering the different objectives, i was thinking of setting up two ceph clusters. Both would be running on the same hardware, the slower would be in a VM with the drive controller passed though to the VM.

The NVMe pool will be replicated to each host and the failure domain is host. This pool is mission critical and maximum speed priority.
The cephfs is be an erasure pool and is not mission critical. The failure domain is osd and speed is low priority.

 

[gurubert]

Don't run Ceph in VMs. Just use the hardware in one cluster.

As the clients exchange data directly with the OSDs there is no bottlebeck where both fast and slow drives would clash (except for the NICs, but these would also be shared in your scenario).

 

[chell]

Don't run Ceph in VMs. Just use the hardware in one cluster.

As the clients exchange data directly with the OSDs there is no bottleneck where both fast and slow drives would clash (except for the NICs, but these would also be shared in your scenario).

Thank you for you reply.
What about memory & cpu management? how can I limit the resources used by the slow pool.

 

[gurubert]

how can I limit the resources used by the slow pool.

I do not think that this will be necessary as the IOPS provided by the NVMe drives will make these OSDs grab CPU and memory resources.
HDD OSD processes usually sit and wait for IO from the HDDs.