On Wed, 1 Sep 2010, H Chang wrote: > Thanks Greg and Sage for your replies. > > Sage's suggestion seems more doable to me. > With custom file layout policy, each file would potentially be able to > use a customized CRUSH rule. > > I see that the feature is planned for v0.22. > Out of curiosity, what will be an interface for specifying such a > policy? Use xattr? That's the current thinking. We haven't settled on anything specific yet. There is also an ioctl to set the layout of an individual file. The xattr value could be that struct, with a name like 'ceph.default_file_layout'. sage > > Best regards, > HS > > On Wed, Sep 1, 2010 at 4:28 PM, Sage Weil <sage@xxxxxxxxxxxx> wrote: > > On Wed, 1 Sep 2010, Gregory Farnum wrote: > > > >> On Wed, Sep 1, 2010 at 11:33 AM, H Chang <heracles@xxxxxxxxx> wrote: > >> > Hi Ceph team, > >> > > >> > I'm a newbie in Ceph, so please bear with me if my question sounds stupid. > >> > > >> > I'd like to explore the possibility to support multiple placement > >> > policies in Ceph. > > > > Some of the infrastructure is there for this. Each file has a layout, > > which specifies the striping strategy and which storage pool the file data > > gets stored. Each pool has a field specifying which CRUSH rule set to use > > (they can share). > > > >> > For example, suppose each user using Ceph has its own placement policy > >> > which may be different from one user to another. > > > > Creating a pool for each user may be overkill, however. The intention was > > for pools and/or crush rules to be defined for things like "fast sas > > disks", "slow sata pool", that sort of thing. Currently everything is > > thrown in a single data pool. > > > > Currently the default file layout is compiled in and applies to all new > > files. There is an open bug/issue for specifying the default file layouts > > on a directory basis (e.g, everthing in /home goes into this pool, > > everything in /scratch goes in that pool). See > > > > http://tracker.newdream.net/issues/185 > > > > In any case, creating per-user pools means pushing more awareness of users > > into the file system, autocreating pools, and that sort of thing. As > > things stand, the layout policy is something an administrator has to > > set up. > > > >> But it will require changing several pretty low-level structs and > >> modifying the metadata server and both versions of the client to > >> enable, so it's not a good introductory project and you shouldn't > >> expect it soon. :( > > > > ...for per-user pools, at least. Something along the lines of issue #185 > > above is more doable (and on the roadmap). > > > > sage > > > >> -Greg > >> -- > >> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > >> the body of a message to majordomo@xxxxxxxxxxxxxxx > >> More majordomo info at http://vger.kernel.org/majordomo-info.html > >> > >> > > > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html > >
