On Sun, Jun 16, 2013 at 11:10 PM, Wolfgang Hennerbichler <wolfgang.hennerbichler@xxxxxxxxxxxxxxxx> wrote: > > > On 06/16/2013 01:27 AM, Matthew Walster wrote: >> In the same way that we have CRUSH maps for determining placement >> groups, I was wondering if anyone had stumbled across a way to influence >> a *client* (be it CephFS or RBD) as to where they should read/write data >> from/to. > > I think the concept of CRUSH doesn't really involve your wish, to write > to specific locations (i was wishing for the same some time in the past, > and then I RTFM'ed more, and in the end found out that this wish is not > very trivial to implement). Although reading locally is possible, as > michael lowe stated in his other e-mail. This is correct. What you can do is set up pools with different rules, such that you have a "West" and "East" pool, you split up your OSDs into "west" and "east" groups, and then the pool selects a primary from the matching set of OSDs and a secondary/tertiary from the other. Then put the RBD images in the appropriate pool. On Sat, Jun 15, 2013 at 6:25 PM, Michael Lowe <j.michael.lowe@xxxxxxxxx> wrote: > My read of http://ceph.com/releases/v0-63-released/ has this for rbd reads > in the dev branch. FYI, right now this really is just *local*, ie you're on the same host. We've had low-intensity discussions around enhancing this core functionality to be a "read-from-closest" model for a while now, but have yet to implement anything because it's much more complex than the simple hack we're currently using and requires stuff like interpreting a closeness model (probably based on CRUSH, but we don't have any similar functionality to borrow from). -Greg Software Engineer #42 @ http://inktank.com | http://ceph.com _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |