Can Ceph Support Multiple Data Centers


A RADOS cluster can theoretically span multiple data centers, with safeguards to ensure data safety. However, replication between Ceph OSDs is synchronous and may lead to low write and recovery performance. When a client writes data to Ceph the primary OSD will not acknowledge the write to the client until the secondary OSDs have written the replicas. See How Ceph Scales for details.

The Ceph community is working to ensure that OSD/monitor heartbeats and peering processes operate effectively with the additional latency that may occur when deploying hardware in different geographic locations. See Monitor/OSD Interaction for details.

If your data centers have dedicated bandwidth and low latency, you can distribute your cluster across data centers easily. If you use a WAN over the Internet, you may need to configure Ceph to ensure effective peering, heartbeat acknowledgement and writes to ensure the cluster performs well with additional WAN latency.

For those who are using the Ceph Object Gateway (RGW) to store objects using the S3 or Swift protocol, there are a sophisticated set of functions and capabilities that allow an administrator to create flexible multi-site topologies by linking, or federating, RGWs. See Configuring Federated Gateways for details.

For those using the Ceph Block Device (RBD), a snapshot capability allows the administrator to create a copy of a volume at a given point in time. This snapshot can be exported as a file or image that can then be imported into a separate Ceph cluster in a different location. See Snapshots for details.