At the current time, Ceph doesn't support overwrite operations on erasure-coded pools and so cannot be directly used for storing RBD images. However, by the use of tiering, an erasure-coded pool can be used as the base pool and a 3x replica pool used as the top level. This allows the hot active dataset which will comprise of frequently accessed data, to sit in the top tier and any infrequently accessed cold data to reside in the base erasure-coded tier.

Care should be taken with this approach; although in theory it sounds like a brilliant idea, it shouldn't be forgotten that in this configuration proxy writes are not supported to the base erasure-coded pool. All write operations to the base tier therefore have to be promoted up to the top tier first. If even a small percentage write operations start targeting your base tier, then performance will quickly fall off a cliff as the constant promotion and flushing operations overwhelm the disks in your cluster. This approach is only recommended for use cases where write operations are infrequent or are contained to a small number of objects.