On 05/13/2013 09:17 AM, w sun wrote:
While planning the usage of fast clone from openstack glance image store to cinder volume, I am a little concerned about possible IO performance impact to the cinder volume service node if I have to perform flattening of the multiple image down the road. Am I right to assume the copying of the blocks incurred by the flattening task is done on the backend nodes of the cluster and there is no extra IO activities on the client side where the librbd operation is issued?
The client reads data from the parent and copies it to the clone. This can't be done by the osds alone since rados does not support multi-object transactions, or having an osd act like a client to another osd right now. There's no limitation on flatten being performed just on one node though. If client I/O is a problem, you can run flattens on as many nodes as you like. Josh _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
