librbd (used by QEMU to provide RBD-backed disks) uses librados and provides the necessary handling for striping across multiple backing objects. When you don't specify "fancy" striping options via "--stripe-count" and "--stripe-unit", it essentially defaults to stripe count of 1 and stripe unit of the object size (defaults to 4MB). The use-case for fancy striping settings for an RBD image are images that have lots of small, sequential IO. The rationale for that is that normally these small, sequential IOs will continue to hit the same PG until the object boundary is crossed. However, if you were to use a small stripe unit that matched your normal IO size (or a small multiple thereof), your small, sequential IO requests would be sent to <stripe count> PGs -- spreading the load. On Wed, Oct 19, 2016 at 12:32 PM, Ahmed Mostafa <ahmedmostafadev@xxxxxxxxx> wrote: > Hello > > From the documentation i understand that clients that uses librados must > perform striping for themselves, but i do not understand how could this be > if we have striping options in ceph ? i mean i can create rbd images that > has configuration for striping, count and unite size. > > So my question is, if i created an RBD image that have striping enabled and > configured, will that make a difference with qemu-rbd ? by difference i mean > enhancing performance of my virtual machines i/o and allowing utilizing the > cluster resources > > Thank you > > _______________________________________________ > ceph-users mailing list > ceph-users@xxxxxxxxxxxxxx > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > -- Jason _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |