When planning a Ceph cluster, consider the following guidelines for the Ceph Monitor and RADOS Gateway services.
The Ceph Monitor service is quorum-based. Three instances of Ceph Monitor are required to ensure fault tolerance and typically suffice for any number of OSD nodes. If additional Ceph Monitors are required for any reason, the total number of instances must be odd.
Run one Ceph Monitor per physical infrastructure node to minimize the risk of losing the Monitor quorum upon server hardware failure.
The recommended minimal size of a Ceph Monitor VM is:
The RADOS Gateway service is required to provide Swift and S3-compatible Object Storage API on top of Ceph. RADOS Gateway is stateless and puts no constraints on the number of nodes in a cluster. Start with two instances behind a load balancer for redundancy in case of failure.
Mirantis recommends running the RADOS Gateway service on a dedicated VM with at least:
RADOS Gateway scales out horizontally, so you can increase the number of VMs to get more concurrent connections processed.