Proxmox Ceph Erasure code Profile change

[sahroni]

dear All
what if we create an erasure code pool, for example I have 10 hosts and I create EC-Profile = k=8m=2 but then later I increase the hosts to 14 hosts, can I edit the EC-profile to k=10m=4,
or is there Another better way to make it more flexible ?

 

[gurubert]

You cannot change the EC profile of an existing pool.
BTW: Your number of hosts should be k+m+2, so with ten host you should use k=6 and m=2.

 

[sahroni]

k+m calculation is the number of OSDs or the number of hosts if I choose domain failure = host? Or is there an easier way to determine it, I don't really understand how to determine the right one

For example, I have 10 nodes and each node has 5 drives, and the drive capacity is 3.8TB, how do I determine the right one if I want to have a lot of capacity but still be safe?

 

[gurubert]

k+m+2 is the minimum number of independent failure zones you have in your cluster.

If your failure zone is on the host level k+m+2 is the minimum number of hosts.

 

[sahroni]

OK, thanks... and what percentage does this get for storage?

and can the default configuration k=2m=1 be used if domain failure = host, with 10 nodes, so that if I add more nodes the configuration is still feasible

 

[gurubert]

k=2 m=1 is not recommended as there is no redundancy left when losing one failure zone.

You can compute the relative usage with the formula (k+m)/k.
With k=6 and m=2 you need 133GB storage space for 100GB of data.

Or the other way around: k/(k+m) means with 100GB storage space you get 75GB for data.

Keep in mind that Ceph fills the OSDs only to about 85% before issuing the first warning, so subtract another 15% from the total storage capacity.

 

[sahroni]

https://openmetal.io/wp-content/upl...-Internet-Erasure-Coding-Calculator-v1.0.xlsx

This wrong ?
10 nodes = k=8m=2

 

[gurubert]

With n being the number of nodes:

n = k + m means as soon as one node is down Ceph does not have any other location to place the data.
n = k + m + 2 gives you enough time in the event of one of the nodes goes down. And it also means that not every node has to participate in every write operation or when recovering data.

 

[sahroni]

Ok thanks