Using ceph for openstack storage
I'm in the process of deploying an Openstack cloud and I've been aiming at maximize the number of computing node and in order to do that I've thought of using Ceph distributed storage and use an All in one openstack deployment, so that swift objects and glance's blocks are stored through Ceph. I have a total of 12 servers, so I'll end up with 1 Controller node for openstack services and use the remaining nodes for ceph cluster and computing. Is it worth it to use a ceph storage system , and what would be the minimum number of nodes required to deploy it.
On 10/31/21 10:54 PM, A Monster wrote:
I'm in the process of deploying an Openstack cloud and I've been aiming at maximize the number of computing node and in order to do that I've thought of using Ceph distributed storage and use an All in one openstack deployment, so that swift objects and glance's blocks are stored through Ceph. I have a total of 12 servers, so I'll end up with 1 Controller node for openstack services and use the remaining nodes for ceph cluster and computing. Is it worth it to use a ceph storage system , and what would be the minimum number of nodes required to deploy it.
Well, are we talking about serious production, or just playing? Since you're talking about a single controller, without any sort of redundancy, I am double-guessing that we're not talking about a serious deployment here. If you were, I would strongly suggest that your Ceph is made of: - 3 ceph mon (that do not share any other role and do only that) - at least 10 ceph OSD nodes, so that loosing one of them doesn't make the full of your cluster super slow (as Ceph automatically would rebalance the data to the 9 remaining nodes). and also that your control plane is made of at least 3 nodes, and not mix Ceph roles with anything else... This would be a serious deployment. But since you seem super limited in therms of hardware, I would suggest that you: - Deploy 3 controllers on which you also install a Ceph MON - Deploy Nova and Ceph OSD on the remaining 9 nodes. This way, you still keep some kind of redundancy at least, if one of the server fails. Remember: server do fail, it's a question of when, rather than if. :) I hope this helps, Cheers, Thomas Goirand (zigo)
participants (2)
-
A Monster
-
Thomas Goirand