On Mon, Apr 6, 2015 at 4:17 AM, Dmitry Meytin <dmitry.meytin@xxxxxxxxxx> wrote: > Hello, > > I want to use CephFS instead of vanilla HDFS. > > I have a question in regards to data locality. > > When I configure the object size (ceph.object.size) as 64MB what will happen > with data striping > (http://ceph.com/docs/master/architecture/#data-striping), is it still will > be striped by unit-size of 64kb? The default striping strategy is just to use 4MB chunks and not to stripe across them at all. If you set your layout to use 64MB chunks and your stripe unit to 64MB (which I think is the default) it will behave the same with 64MB of contiguous data on a node. So there's no problem. -Greg > If so, the whole meaning of the data locality will be lost, since the data > will be pulverized across an entire cluster. > > What are the best practice to deal with this problem (Ceph is used for many > different purposes, including ephemeral storage, elastic block storage and > image repository – entire OpenStack storage solution) > > > > Thanks, > > Dmitry > > > _______________________________________________ > ceph-users mailing list > ceph-users@xxxxxxxxxxxxxx > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |