Proxmox VE clusters
Initial design, three-node clusters with HA failover, Ceph or ZFS storage, automated backup with PBS. We have done VMware-to-Proxmox migrations end-to-end with rollback plans and verified restores.
— Practice / Infrastructure
Some workloads belong on the edge. Some belong on bare metal. We work both halves, and we treat the second one with the same engineering rigour as the first — Proxmox clusters, Linux servers, hardening, observability, and the boring operational discipline that keeps systems running for years instead of quarters.
— What we ship
Initial design, three-node clusters with HA failover, Ceph or ZFS storage, automated backup with PBS. We have done VMware-to-Proxmox migrations end-to-end with rollback plans and verified restores.
New colocation deployments, Hetzner / OVH / Latitude.sh dedicated machines, IPMI configuration, network segmentation. Proper DNS, proper firewalling, proper documentation.
CIS Benchmarks-aligned hardening: SSH key-only auth, fail2ban, unattended-upgrades, ufw / nftables, disabled services, audit logging. Every server gets the same baseline before workload deploy.
Gitea / Forgejo, Drone / Woodpecker CI, container registries, Vaultwarden, Nextcloud — the open-source equivalents of GitHub / GitLab CI / 1Password / Dropbox, run on your hardware.
Borg / restic for filesystem backups, Proxmox Backup Server for VMs, off-site replication. Prometheus + Grafana + Alertmanager or Uptime Kuma — sized to the org, not the catalog.
When the right answer is some-of-both — Cloudflare front, on-prem origin, secure tunnel between them — we design and operate the whole topology. Cloudflare Tunnel, WireGuard, Tailscale, mTLS. No public IPs, no NAT pain.
— Why this practice exists
A 64-core Proxmox node with 256 GB RAM at Hetzner is CA $200/month. Equivalent compute on AWS or GCP is 5 – 10× that, before egress. For workloads that don't need elasticity, on-prem or dedicated wins by a wide margin — provided someone competent runs it.
For regulated workloads, customer-data residency, or businesses that don't want to be a tenant on a hyperscaler, owning the substrate is the answer. PIPEDA-sensitive Canadian data, healthcare records, financial system-of-record — these belong somewhere you can audit and govern directly.
Most teams that run their own servers learned to do it five years ago and haven't iterated. The result: stale Debian releases, manual deploys, no monitoring, backups that have never been restored. We fix that — bring the systems to a baseline that survives the next decade.
— Stack
Hypervisor
Proxmox VE · KVM · LXC
OS
Debian · Ubuntu LTS · Alpine · Rocky
Provisioning
Ansible · cloud-init · Terraform (where it fits)
Networking
WireGuard · Tailscale · Cloudflare Tunnel · pfSense / OPNsense
Storage
ZFS · Ceph · LVM · NFS · S3-compat (MinIO / Garage)
Backups
Proxmox Backup Server · Borg · restic · rsnapshot
Observability
Prometheus · Grafana · Loki · Alertmanager · Uptime Kuma
Containers
Docker · Podman · Compose · Nomad (for fleets)
Edge handoff
Cloudflare Tunnel · Caddy · nginx · Traefik
Hosts
Hetzner · OVH · Latitude.sh · in-house
— Best fit
— Not a fit
— See also
— Read deeper
Ransomware hunts your backups before it encrypts your files. Here is the 3-2-1 rule extended to 3-2-1-1-0 — offline and immutable copies, retention that outlasts an attacker's patience, and why the online backup you trust is the one they delete first.
Self-hosting on Proxmox or Hetzner has a real sticker-price advantage and a real time cost. Managed cloud inverts both. Here is an honest comparison of the hidden costs on each side and a framework for deciding.
How Proxmox HA actually works — cluster, quorum, replicated storage, fencing, and automatic restart — what it protects against, what it does not, and when the complexity is worth it for a small business.
Register a Windows device into Intune and Autopilot as corporate with a single command at OOBE — how our published autopilot script works, the enrollment traps it avoids, and how to harden an irm | iex supply chain.
Two years after Broadcom completed the VMware acquisition, the migration patterns are clear. Proxmox VE leads on small-to-mid fleets. XCP-ng and Nutanix split the higher-end market. Here is the 2026 decision framework, with real cost figures.
Tell us the shape of the workload, the constraint, and the deadline. We respond in writing within one business day.