I'm planning a Proxmox cluster built around two identical Dell PowerEdge R650 servers, with a third standard PC used for quorum. The environment will host roughly 30–40 Windows and Linux virtual machines, along with a CCTV NVR workload. Before committing to this substantial purchase, I'd appreciate feedback on potential bottlenecks, compatibility concerns, missing components, or ways to improve the design.
Each R650 is configured with 2× Intel Xeon Platinum 8380 processors, 384 GB of DDR4-3200 RDIMM memory using 12× 32 GB modules, a Dell HBA345 front storage controller, an M.2 NVMe HBA with two 1 TB NVMe SSDs, four 10GbE SFP+ ports through a Dell QLogic QL41154 OCP 3.0 adapter, dual 800 W Platinum power supplies, and iDRAC 9 Enterprise. Both hosts are intended to use identical hardware. I'm also considering how best to separate storage, management, and VM traffic, how to arrange the storage volumes for future expansion, and what backup strategy should accompany the cluster. Any practical advice before ordering would be very welcome.
3 Answers
The overall design looks capable, especially if both R650s are configured identically. Make sure you plan separate networks or VLANs for management, VM traffic, storage, and cluster communication where possible. Also, don’t treat RAID or VM replication as a backup—keep independent backups and test restoring them. The third machine can help with quorum, but it should be reliable and connected to a stable network.
Before buying this much hardware, compare the total cost and operational effort with a managed private-cloud provider. For a few dozen always-on VMs, hosting them externally may be cheaper once power, support, replacement parts, backups, maintenance, and your own time are included. If you specifically need on-premises control or have compliance, latency, or CCTV requirements that rule out hosted infrastructure, then the local cluster can still make sense—but it’s worth pricing both options honestly.
Leave room for storage growth when laying out the volumes. In particular, consider how the initial system or data volume placement will affect later expansion, and document the layout before deploying workloads. It’s much easier to plan the expansion path now than to reorganize storage after the VMs are running.
That’s helpful—I’ll account for future expansion in the volume layout before building the cluster.
Both R650s are planned to be identical. I’ll also revisit the network layout and make sure there’s a proper independent backup plan rather than relying only on the cluster storage.