CEPH equivalent configuration for multiple hosts with local RAID

[tawh]

I set a 3 proxmox hosts cluster and each host has 1x10TB HDD, CEPH configured with replication mode so effective storage in cluster is 10TB.
As the IO performance is poor, I am thinking to replace each 10TB disks with 3x6TB disks, I want maintain the host redundancy level like before but enable the RAID-5 like storage system in each host. After the replacement, the effective cluster storage is 2x6TB=12TB.

In CEPH, there is a Erasure code config with lRC plugin (https://docs.ceph.com/en/latest/rados/operations/erasure-code-lrc/), I am wondering if the
----------------------------------------------
k=2
m=1
l=3
crush-locality=host
crush-failure-domain=host
----------------------------------------------
is equivalent to have a cluster storage with replication among all hosts and each host has a RAID-5 like protection. Thanks

 

[Alwin]

k=2 m=1 l=3

Only one OSD can die for a data chunk. Any subsequent failure may result in the loss of that chunk. Also the k+m needs to be a multiple of l.

In such a small setup, you will not gain much on space, let alone performance. Best stick with the 3/2 pool size and add more OSDs. This way you can be sure that your data is save in case of a node failure.

 

[tawh]

Only one OSD can die for a data chunk. Any subsequent failure may result in the loss of that chunk. Also the k+m needs to be a multiple of l.

In such a small setup, you will not gain much on space, let alone performance. Best stick with the 3/2 pool size and add more OSDs. This way you can be sure that your data is save in case of a node failure.

Thanks for your reply.
To verify the correctness of my understanding, If I deploy this policy to 3 hosts and form a pool with 9 OSDs (as mentioned in Post #1), will CEPH replicate data to the other 2 hosts? or it just pick an arbitrary host to store the data?

 

[Alwin]

To verify the correctness of my understanding, If I deploy this policy to 3 hosts and form a pool with 9 OSDs (as mentioned in Post #1), will CEPH replicate data to the other 2 hosts? or it just pick an arbitrary host to store the data?

The policy will be set on pool level, including all hosts participating. As the crush locality & failure-domain are on host, each chunk set will be placed on a different node.

Though still, my above warning stands.