#MySQL on Ceph using Rados
summary
Prototype implements MySQL storing InnoDB pages on Ceph via RADOS and RBD, using snapshot+clone provisioning to spin slaves that share a base image (demo shows master using ~311MB and clones ~50MB). Tested on 48 OSD spinner cluster yielding ~100 IOPS per disk and single-instance peaks >10k IOPS; network latency matches iSCSI. Code compiles with a segfault in SEF, tools and scripts live on GitHub, and parallel doublewrite buffer materially raises throughput. RADOS block devices (RBD)
AI-written summary · about
chapters (24)
- 0:00Intro
- 0:26Who am I?
- 0:57Plan
- 1:42Ceph 101: general
- 2:57Ceph 101: versions
- 3:39Ceph 101: nodes/processes
- 4:24Ceph 101: OSD nodes
- 5:48Ceph 101: MON
- 7:51Ceph 101: how objects are stored
- 9:06Ceph 101: pools
- 10:49Ceph 101: access
- 12:59A Ceph 101: rbd
- 14:42A Ceph 101: CephFS
- 15:13Ceph 101: Deploying
- 16:11Ceph 101: My home Ceph cluster
- 16:49Ceph brings to MySQL: the basics.
- 18:02Ceph brings to MySQL: Efficient storage
- 18:41Ceph brings to MySQL: thin prov. slaves(2)
- 28:53Ceph brings to MySQL: thin prov. Slaves(3)
- 31:44Ceph brings to MySQL: thin prov. PXC
- 33:13MySQL/Rados: the idea
- 39:15MySQL/Rados: RadosFS
- 39:53MySQL/Rados: Modifying InnoDB
- 42:18MySQL/Rados: Status
related talks
48mCould Ceph Become More Like a Database?
20mExtreme Performance and SmartScaling MySQL with Storage Class Memory Servers
29mCeph Object Storage Overview, Capabilities and Future Plans
22mMySQL and OpenStack
55mThe Future of MySQL
58m600k QPS on MySQL Galera cluster
41mDemocratizing MySQL: From Cloud Managed to Kubernetes Managed
34mEvaluating the Performance and Power Efficiency of Ceph Storage Configurations