Ceph hardware sizing: CPU, memory, drives and networks
Turning workload numbers into a bill of materials: drives per host, memory per OSD at peak, CPU for erasure coding and RGW, bonded networks, and benchmarking candidate hardware before buying.
Ceph cluster design and deployment
Design and deployment decisions for a new Ceph cluster: hardware, placement groups, failure domains, media, monitors, networks, capacity and cephadm deployment, including air-gapped sites.
Setting Ceph placement group counts and the PG autoscaler
Getting PG counts right from the start: autoscaler modes, target_size_ratio for pools that will grow, bulk pools, mon_target_pg_per_osd and reading autoscale-status recommendations.
Designing Ceph failure domains across racks and rooms
Choosing host, rack, row or room as the failure domain: setting CRUSH locations, how many failure domains a replica count or erasure code profile needs, and what it means for usable capacity.
Mixing HDD, SSD and NVMe in Ceph with device classes
Separating fast and slow media into different pools: automatic device classes, class-specific CRUSH rules, hybrid rules with SSD primaries and the read penalty when a primary SSD fails.
Sizing BlueStore DB and WAL devices for HDD OSDs
How much SSD to put in front of each HDD OSD: the 2.5% rule from Squid onward, older RGW and RBD guidance, spillover when the DB fills, and how many OSDs to share one SSD.
How many Ceph monitors you need and where to place them
Three versus five monitors, odd numbers for quorum, spreading monitors across failure domains, monitor disk space warnings and clock synchronization.
Separate public and cluster networks for Ceph
Whether to split client and replication traffic: the cluster network option at bootstrap, bandwidth during recovery, bonding across two switches and the msgr2 ports to allow.
Ceph capacity planning: usable capacity and full ratios
Working out usable capacity from raw disks: replication or erasure coding overhead, headroom for a failed host, nearfull, backfillfull and full ratios, and when to order more hardware.
Designing a three-node Ceph cluster
What a minimal production cluster can and cannot do: three monitors, host failure domain with three replicas, recovery when one node is down, erasure coding limits and when to start with more nodes.
All-flash NVMe Ceph cluster design
Designing for low latency: enterprise NVMe drives, CPU and memory per OSD, networks that keep up with flash, and whether separate DB devices still help.
Bootstrapping a production Ceph cluster with cephadm
From bare hosts to a healthy cluster: host requirements, bootstrap options, adding hosts over SSH, admin labels, placing monitors and managers, and checks before handing over.
Deploying Ceph in an air-gapped network with cephadm
Building a cluster with no internet access: a local registry for Ceph and monitoring images, insecure versus TLS registries, bootstrap with a local image and keeping the mirror current for upgrades.
Deploying Ceph OSDs with cephadm service specifications
Describing drive layouts declaratively: data and DB device filters by rotational, size and model, dry runs, all-available-devices versus explicit specs, unmanaged mode and encrypted OSDs.
