Tuning osd_memory_target and BlueStore cache
Sizing OSD memory to the host: the 4 GiB default, autotuning, minimums, how cache size affects metadata-heavy workloads and avoiding swap or OOM kills.
Ceph performance tuning
Tuning a running Ceph cluster: OSD memory, recovery versus client I/O, benchmarking, compression and data balance.
Balancing recovery and client I/O with the Ceph mClock scheduler
Choosing a profile when a failed drive slows applications or recovery drags on: balanced, high_client_ops and high_recovery_ops, OSD capacity benchmarks and custom settings.
Benchmarking a Ceph cluster before production
Measuring what the hardware can deliver: drive and network baselines, OSD capacity benchmarks, pool-level tests that match the workload and recording results to compare after changes.
BlueStore compression for Ceph pools
When compression saves capacity and when it costs CPU for nothing: algorithms, modes, the required ratio, blob sizes and pools that suit compression.
Balancing data across OSDs with the upmap and read balancers
Fixing uneven OSD use: upmap mode, misplaced ratio limits during balancing, the deviation target and read balancing for primary PGs on Reef and later clients.
