is this a correct number of pg placement in datapool ?

[FSNaval]

Good day all,

I have a three node proxmox cluster running ceph and i have created two pools.

The first pool (vmpool) consisted from NVME drives for my VMs which ceph assigned 128pgs

The second pool (datapool) consisted from HDD drives for my VMs which ceph assigned 32pgs

Please see attached image. On both pools, pg assignment has been done automatically and for both pools as you can see the "PG Autoscale Mode" is "ON"

I think that the number of PGs on the datapool is low, how it is possible to have a lower number despite having bigger capacity than (vmpool)? should i increase the number of PGs manually? What is your opinion?

 

[Falk R.]

The automatic calculation of the optimum PGs depends on the number of OSDs and the amount of data in the pool. As the pool is still empty, this results in a small number of PGs.
It is best to check how much data should be in the pool and calculate this manually using a PG calculator.

 

[spirit]

if you have only 1 pool for each kind of disk, set target ratio to "1.0"

 

[FSNaval]

The automatic calculation of the optimum PGs depends on the number of OSDs and the amount of data in the pool. As the pool is still empty, this results in a small number of PGs.
It is best to check how much data should be in the pool and calculate this manually using a PG calculator.

Thank you very much for your prompt reply.

I have used this pg calculator: https://docs.ceph.com/en/latest/rados/operations/pgcalc/

my pool is the last one (my homedatapool) and as per calculations it should have 128PGs instead of 32 that it currently has.

I used a 3-node cluster with 10 OSDs totally and an expected capacity of 30% of data.

How can i verify this? Should i proceed with manually adjusting the PGs?

 

[Falk R.]

Yes, do that. 128 PG sounds plausible and it can be changed at any time. Then it only has a performance impact.

 

[FSNaval]

if you have only 1 pool for each kind of disk, set target ratio to "1.0"

Yes, do that. 128 PG sounds plausible and it can be changed at any time. Then it only has a performance impact.

I have read in the documentation that between 5 and 10 OSDs i should set pg_num to 512. Using the calculation relation from the documentation, the nearest power of two is 512.

How much a performance impact there should be between 512 and 128?

Maybe i should opt for 256 instead and see how it goes?

 

[Falk R.]

How much a performance impact there should be between 512 and 128?

It always depends on the data that is stored on it. The difference is guaranteed to be measurable, but you won't usually feel it.