Its possible to control where nova stores ephemeral volumes using the "nova_instance_datadir_volume" variable. All you need to do is provide it a mount point on the HV and it will bind it into the libvirt container:
So provided you can raid together the NVME drives it should work.
CAUTION:
This email originates from outside THG
Hello.
I would like your opinion regarding a hardware configuration and my Openstack cluster project.
I have the possibility of purchasing 6 servers to build this cluster for students, my current Openstack being too small.
I have a proposition from the manufacturer (DELL): 6 servers with characteristics per server:
- BOSS-N1 controller card + with 2 M.2 960GB (RAID 1)
- 5 x 3.84TB Data Center NVMe Read Intensive AG Drive U2 Gen4 with Carrier
- 12 x 64GB RDIMM, 4800MT/s Dual Rank
My problem is: BOSS is used to install the system (Ubuntu 22.04 in my case).
But, various tests on a small cluster with SSD disks showed that the solution with ephemerals was really the most interesting for what I want to do.
Problem: I don't believe you can have the system on the BOSS and the ephemerals on the NVMe disks.
So my questions are:
1) am I wrong and ephemerals are possible on NVMe (I use kolla-ansible to deploy, I haven't seen how to do that)?
2) or the solution is to create a Ceph cluster on the NVME of the 6 servers
3) or a better solution exists
Thank you in advance if you can help me make these choices.