Feature #23505
rbd zero copy
0%
Description
rbd zero copy would just ask the Ceph cluster to "copy $rbd image" to a new image. This saves a lot of bandwith back and forth between the client and the cluster (OSDs). In theory, especially on clients with limitited network throughput, this would mean incredible speed improvements. This would allow for "fastest rbd copy" the cluster could possible peform. I guess, behind the scenes, rbd (zero) copy is "just" a matter of "duplicating" the objects referenced by an rbd image to new unique objects and registering them as an rbd image.
Note: "zero" copy is implemented in some commercial filers, where i.e. cloud management controllers can tell the filer to "zero copy" a "golden image" to a new image on it's behalf.
History
#1 Updated by Jason Dillaman almost 6 years ago
This is basically implemented via "rbd clone".
#2 Updated by Stefan Kooman almost 6 years ago
Hi Jason,
you mean by the following commands?
rbd snap create
rbd snap protect
rbd clone
rbd flatten
rbd snap unprotect
rbd rm snap
My idea was to make it very simple and fast, and without the need for a lot of rbd operations / bookkeeping. If it's already "basically implemented", would it be a lot of work to make a new rbd command?
#3 Updated by Jason Dillaman about 4 years ago
- Status changed from New to Rejected