Rebuilding the HVENS Cloud Network from the Ground Up
Starting in 2023, we stopped iterating on the legacy HVENS network and redesigned it from scratch. This is the short version of what changed and why.
The problem with "just keep patching it"
Our original cloud fabric had grown organically over a decade. VLAN-based tenant separation worked fine for the workloads we had in 2015 - mostly single-tenant ISP-adjacent services. As MSP customers started asking us to run real multi-tenant workloads with tighter isolation and cleaner failover, patching a VLAN topology to behave like something modern stopped being worth the effort.
What the new design looks like
- EVPN for tenant segmentation. Each customer lives in their own VRF/EVPN instance. No more "which VLAN ID is free this week" spreadsheet.
- Two sites, one fabric. Ashland (Pixel Factory, VA) is production. Denton (Choptank Fiber, MD) is the replication and DR target. Separate power grids, separate fiber paths, ~150 miles apart.
- Proxmox VE as the hypervisor standard. We've been a Proxmox partner for years; standardizing everything on Proxmox VE + PBS made the operational story drastically simpler.
- Storage tiers we actually mean. Ceph for general-purpose VMs, ZFS for workloads that prefer local-attached performance, Blockbridge SAN for customers that need dedicated performance profiles.
What we kept
The things that were already working - DE-CIX peering, direct peering with Netflix/Google/Akamai, carrier-neutral transit out of Pixel Factory - didn't need to change. The redesign was the fabric and the tenant model, not the upstream connectivity.
Why this matters for customers
Most customers never see the fabric. What they do see: cleaner tenant isolation, predictable DR behavior, and an operations team that can answer a 2am question about their VM without a 20-minute archaeology dig through VLAN configs. That's the part we were optimizing for.
If you're curious about specifics - EVPN route-target design, how we handle Ceph placement groups across the two sites, or how PBS offsite retention is wired up - get in touch. We're happy to go deep.
Related
- HVENS infrastructure - the two PoPs and the network described in this post.
- Proxmox at HVENS - we're a Silver-level Proxmox Partner; the cloud is built on PVE.
- Cloud catalog - compute, storage, connectivity, and Backup & DR.