Purchasing Hardware for Ceph Cluster

[cghess]

Currently going through the process of working with a Dell sales team to buy hardware for a production environment. Right now, I believe the VMs hosted on my initial build will be basic microsoft domain controllers, file server, and a few more basic windows vm for licensing.

In regards to Ceph cluster, their documentation prioritizes single thread performance which would have me favor Intel CPUs for this build, although I personally would prefer AMD. For Ceph, what is the community consensus for processors?

For the memory, because I have no plans on hosting hundreds of VMs at this time, but would like to be comfortable if more needed to be added in the future. I believe 128 GB would be plenty moving forward, am I correct with that assumption?

Lastly, storage that was advised from Dell was 2x 1.6 TB NVMe, with my understanding of Ceph, this would be a comfortable amount as I would be able to set one of the drives as a part of the ceph pool.

Network wise, looking to go for 25GbE between the nodes. Relatively new, however I am trying to advocate for full proxmox migration in our current environment so any input or concerns would be greatly appreciated!

Thank you!

 

[alexskysilk]

their documentation prioritizes single thread performance which would have me favor Intel CPUs for this build, although I personally would prefer AMD. For Ceph, what is the community consensus for processors?

the consequences of this arent really impacted by the TYPE of cpu you use. AMD and intel are both fine for ceph.

For the memory, because I have no plans on hosting hundreds of VMs at this time, but would like to be comfortable if more needed to be added in the future. I believe 128 GB would be plenty moving forward, am I correct with that assumption?

Many variables aside of number of vms. how many nodes, OSDs, etc.

Lastly, storage that was advised from Dell was 2x 1.6 TB NVMe, with my understanding of Ceph, this would be a comfortable amount as I would be able to set one of the drives as a part of the ceph pool.

I... dont think you and your dell rep were speaking the same language. that, or he doesnt know anything about ceph and was just pitching you standard loadout. Lets go back to the beginning:
- What is your minimum required space? do you intend to deploy this for both virtual storage and other functionality (eg, images, isos, filespace, etc.)
- What is your minimum required IOPs for the entire filesystem (across all guests)
- How many nodes are you planning to implement?

Network wise, looking to go for 25GbE between the nodes. Relatively new, however I am trying to advocate for full proxmox migration in our current environment so any input or concerns would be greatly appreciated!

25gb is probably fine- but you should already be thinking of multiple links for failover, corosync, service network separation, etc. this is valid for any and all clustered environments.