Ceph

Sage Weil’s 2007 PhD thesis topic at University of California, Santa Cruz.

The service Ceph provides is RADOS, Reliable Autonomic Distributed Object Store.

RADOS provides four services: LIBRADOS, RADOSGW, RBD, and CEPHFS.

RADOS

Image courtesy of Sage Weil and Lisa

OSD

OSD — Object Storage Daemon

ceph-osd

Manages the disks and data
1 per HDD or SSD
Responsible for serving data
Cooperatively peers, replicates, rebalances data
1-1000+ per cluster

MON

Mon — Monitor

ceph-mon

Central authority for authentication, data placement, policy
Coordination point for all other cluster components
3-7 per cluster

MGR

Mgr — Manager

ceph-mgr

Host for pluggable management functions
Metrics
- throughput, disk usage, etc.
1 active, 1+ standby per cluster

MDS

ceph-mds

Manage file system namespace
Store file system metadata in RADOS objects
- File and directory metadata (names, inodes)
Coordinate file access between clients
Manage client cache consistency, locks, leases
Not in dataplane
1s - 10s active, plus standbys

RADOS cluster

ceph-cluster

References

Ceph: Reliable, Scalable, and High-Performance Distributed Storage - Sage Weil

Ceph: Managing a distributed storage system at scale - Sage Weil & Lisa - Inktank

CephFS: Architecture Introduction & New Features - Greg Farnum

Ceph Tech Talk - Intro to Ceph - 2019-JUN-27 - Sage Weil - YouTube

Ceph Tech Talk - Intro to Ceph - 2019-JUN-27 - Sage Weil - Slides