New deployment hardware specs

[cmeyer0111]

Hello,

I am sure this has been asked 100 different ways, but I am brand new to ProxMox and have been asked to setup a new deployment. We need to migrate 9 VMs from a VMware deployment. We have primarily used Hyper-V clusters or Azure in the past, however, a few of the servers are running some older Linux versions that won't run on these hypervisors. (Ideally these VMs would have been upgraded years ago, but you deal with what you are given sometimes.) I have been able to boot them on ProxMox in a lab environment already. And now we are looking to build out a HA cluster deployment to run these machines. As currently deployed the machines are allocated 26vCPUs 106 GB RAM and 4.35 TB worth of VM disks. However, the workloads on the servers vary. There is a DB on one, and pair of RDS servers to connect to said DB. There is also a Domain controller for the RDS servers. As well as a few other applications that facilitate file uploads and storage sprinkled on the rest

We are considering 2 or 3 Dell R360 servers with 2.6Ghz 8C/16T CPUs, 128GB RAM, 480GB BOSS Card (for ProxMox OS), 8 x 4TB SATA Drives w/o RAID card (I was reading we could use Ceph to directly cluster physical disks). I am still reading and learning exactly how ProxMox does HA clusters with Hyperconverged nodes and am leaning towards 3 nodes but wanted to hear others experience. My main question is whether the processor here is sufficient to cover the vCPUs as currently allocated. Would it be better to get 3 nodes with lower CPU specs or 2 nodes and increase the CPU core/thread count?

Any other problems you see, pointers you want to share or guidance you can give would be tremendously appreciated.

Cheers

 

[sterzy]

My main question is whether the processor here is sufficient to cover the vCPUs as currently allocated. Would it be better to get 3 nodes with lower CPU specs or 2 nodes and increase the CPU core/thread count?

Well if you want proper fault tolerance of a node, I'd highly recommend three nodes as a three node cluster can handle the loss of a single node much better. Both in-terms of the way that Corosync clustering works and also, because the HA resources can be distributed better.

Think of it this way: If you only have two nodes and one fails, the other needs to be able 100% of the load of the entire cluster. If you have three nodes and one fails, the other two only need to handle about 50% of the load of the entire cluster. So you can lower the specs a single node needs to handle the load somewhat.

 

[cmeyer0111]

Well if you want proper fault tolerance of a node, I'd highly recommend three nodes as a three node cluster can handle the loss of a single node much better. Both in-terms of the way that Corosync clustering works and also, because the HA resources can be distributed better.

Think of it this way: If you only have two nodes and one fails, the other needs to be able 100% of the load of the entire cluster. If you have three nodes and one fails, the other two only need to handle about 50% of the load of the entire cluster. So you can lower the specs a single node needs to handle the load somewhat.

Thank you for clarifying. I suspected this was the case but like I said I am really new to ProxMox.

As far as the vCPU count vs core/thread count on the physical CPUs does the ratio look correct or maybe even over provisioned?