Start with a failure scenario
Choose a concrete event: a failed drive, an unavailable server, or loss of a rack. A CRUSH rule describes where copies or coding chunks may be placed. The physical network and power layout must agree with the failure domains described in the configuration.
Separate data durability from service availability
Enough surviving copies are necessary, but do not alone guarantee that a client can keep working. Monitor quorum, pool settings, reachable gateways and client reconnection behaviour also affect availability. Test the application path, not only cluster health.
Size for recovery as well as normal I/O
A recovering cluster must serve clients while moving data. Budget CPU, memory, network capacity and free disk space for that state. Dense nodes can make the impact of a single host failure much larger. Test representative workloads in both healthy and degraded conditions.
Choose an operating model
Dedicated storage lets compute and storage grow independently. A hyperconverged design shares server resources with workloads and needs careful resource planning. Neither model is universally better: ownership, growth and recovery requirements determine the choice.
EURONAS SOLUTION
Explore euroNAS eEKAS
euroNAS eEKAS brings Ceph-based storage to IT teams through graphical setup and management, with integrated block, file and object services.
Discover eEKAS at euroNAS ↗