Memory requirements

The overall amount of random access memory (RAM) is dictated by the number of disks and overall size of the cluster. Each service type has its minimum memory amount. When building a server, the total memory must be calculated based on the number of services collocated on it (e.g. 12 disks equals 12 OSDs, each requiring a minimum RAM amount), and additionally the operating system itself also needs to be considered.

The table below specifies memory requirements for different service types, and the Ubuntu OS:

OS/Ceph service	Memory requirements (GB)	Performance notes
Ubuntu	16
MON	8	RAM consumption increases with cluster instability.
OSD	8	HDD requires a minimum of 4 GB, but will perform better at 8 GB. SSD and NVMe require more memory, consuming up to 16–24 GB per disk depending on tuning.
MDS	64	Up to 96 GB can improve performance; after which horizontal scaling is recommended.
RadosGW	4	Up to 32 GB can improve performance; after that, horizontal scaling is recommended.
NVMe-oF GW	8	8 GB minimum per daemon. Requires more memory proportional to the number of RBD images mapped and IOPS. 16+ GB is required for 200,000+ IOPS scenarios.

Note

Recommendations will change with the size and workload of the cluster.