Ceph Object Storage Metadata Reduction
Overview
In Ceph object storage, to reduce data amplification at the backend, the data pool adopts the EC mode instead of replication. In EC mode, the metadata amplification factor is the number of EC copies, and metadata amplification is severe. This feature extracts xattr metadata of RGW objects from the data pool into a new pool adopting the replication mode to reduce metadata amplification.
Technical Principles
Metadata stored in EC mode is changed to the replication mode to reduce metadata write amplification. Ceph RGW object metadata can be stored in NVMe SSDs and will not overflow to HDDs, preventing performance deterioration caused by metadata overflow.
Expected Results
Compared with the solution before metadata separation, metadata (excluding object maps) is reduced by 15% under the same data volume. In extreme conditions with 10 billion objects, the read/write performance decreases by no more than 50% compared with that of 100 million objects (100 KB block size).