On Fri, Jan 18, 2019 at 12:42:21PM +0100, Robert Sander wrote: > On 18.01.19 11:48, Eugen Leitl wrote: > > > OSD on every node (Bluestore), journal on SSD (do I need a directory, or a dedicated partition? How large, assuming 2 TB and 4 TB Bluestore HDDs?) > > You need a partition on the SSD for the block.db (it's not a journal Thanks, didn't realize that. Can I do that and remain flexible by going LVM on both SDD and HDD? > anymore with blustore). You should look into osd_memory_target to > configure the osd process with 1 or 2 GB of RAM in your setup. Got that. > > Can I run ceph-mon instances on the two D510, or would that already overload them? No sense to try running 2x monitors on D510 and one on the 330, right? > > Yes, Mons need some resources. If you have set osd_memory_target they > may fit on your Atoms. Thanks. I'll guess I'll have to experiment a little. It's not that I'll be using this heavily. > > I've just realized that I'll also need ceph-mgr daemons on the hosts running ceph-mon. I don't see the added system resource requirements for these. > > The mgr process is quite light in resource usage. Good to know. > > Assuming BlueStore is too fat for my crappy nodes, do I need to go to FileStore? If yes, then with xfs as the file system? Journal on the SSD as a directory, then? > > Journal for FileStore is also a block device. Didn't realize that, either. Thank you for your answers. Appreciated. _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |