the FrankeNAS - (Raspberry Pi, Zima Board, Dell Server, Ugreen) // a CEPH Tutorial

NetworkChuck Technology Careers Career & Money

About this video

The creator builds a ‘FrankeNAS’ from spare hardware — laptops, Raspberry Pis, a ZimaBoard, UGREEN and TerraMaster NAS units, and an old Dell server — using Ceph (rendered ‘Seth/SEF’ throughout the transcript), open-source software-defined storage that is hardware-agnostic. He contrasts this with his four-server production cluster ‘Hagrid,’ built after a 16-bay Synology filled up: separate NAS boxes cannot combine, while Ceph scales by adding any servers. Ceph architecture is explained in depth: a manager/monitor host and additional hosts form a cluster, with monitors forming a quorum; each individual drive becomes an OSD (object storage daemon) handling storage, replication, recovery, and rebalancing; RADOS writes files as objects into pools (e.g. an SSD-only ‘current projects’ pool using replication, and an archive pool using erasure coding, likened to RAID 5/6); placement groups (PGs) sit between pools and OSDs, each PG assigned to three OSDs with a primary that replicates; the CRUSH algorithm and crush map/rules determine placement and rebalancing without a central database; and CephFS separates metadata onto MDS (metadata server) roles. The tutorial preps Ubuntu hosts (22.04 generally, 20.04 on Raspberry Pi for Ceph Reef 18.2.0.2), SSH/root access, wiped drives, Docker, LVM2, and time sync, then uses cephadm to bootstrap, copies SSH keys, adds hosts with ‘ceph orch host add’, adds all available drives with ‘ceph orch apply osd –all-available-devices’, and creates a filesystem with ‘ceph fs volume create’. His result, as shown: 7 hosts, 22 OSDs, 38 terabytes, and 529 placement groups at health OK. He mounts CephFS natively on Linux, shares it via Samba/SMB to Windows (about 111 MB/s in his test), and notes his production cluster also uses tiering, block storage, and Proxmox integration.

Key takeaways

Ceph is open-source, hardware-agnostic software-defined storage that pools mismatched drives and servers into one cluster
Each drive is an OSD; monitors form a quorum, pools set replication or erasure coding, and PGs map objects to OSDs
CRUSH decides object placement and rebalancing algorithmically; CephFS offloads metadata to MDS roles
cephadm workflow: bootstrap one host, add hosts with ceph orch, then apply all available devices as OSDs
His lab result as shown: 7 hosts, 22 OSDs, 38 TB, health OK; access via native Linux mount or Samba/SMB

Episode details: published July 26, 2024 · 49:25 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are NetworkChuck’s own claims, shared for education — they are not typical or guaranteed.

← Previous: Linux on Windows……Windows on Linux · Next: I forced EVERYONE to use Linux →

All NetworkChuck videos · NetworkChuck profile · NetworkChuck on YouTube