ceph how work

asked 2019-12-13 03:37:04 -0500

lelunicu gravatar image

hi, i like to understand how cephs work.

let`s say we have 4 ceph server storage and at a certain moment a node write a file in ceph.then 4k from this file is written in ceph server 1,the next 4k in server 2,so on,finnaly the ceph servers sync together and the file is written in every ceph server.

ceph server is a parallel storage? tnhx

edit retag flag offensive close merge delete


Read the docs for more detailed information. Basically, one client writes to multiple placement groups (PG) at the same time, each PG and its replicas lies on multiple OSDs (disks). So the client writes to multiple OSDs simultaneously.

eblock gravatar imageeblock ( 2019-12-13 08:03:45 -0500 )edit

That makes it slower for one client, but because of the scalability it's a lot faster for many parallel clients. Each PG has a primary OSD which replicates its data to the other OSDs when writing, the "ack" happens when all replicas have been written. Reads are much faster (only primary PG is read).

eblock gravatar imageeblock ( 2019-12-13 08:07:25 -0500 )edit

this mean that if a worker write a file then this file has the same content in every ceph storage server.right?

lelunicu gravatar imagelelunicu ( 2019-12-13 09:29:07 -0500 )edit

The file is split into multiple stripes, these stripes are replicated across your cluster and are identical.

eblock gravatar imageeblock ( 2019-12-16 03:07:18 -0500 )edit