Everything above the network inherits its problems. A storage cluster is only as reliable as the fabric between its nodes, and a virtualization platform is only as flexible as the overlay it can build. So we treat the network as infrastructure to be designed, not cabling to be terminated.
Where our experience comes from
We built a working VXLAN mesh before EVPN was a practical option. There was no control plane to lean on, so we wrote one: a key/value store distributing endpoint information, feeding the Linux bridge forwarding database directly.
That is not the way to build it today โ EVPN exists now and we use it. But having built the layer underneath by hand means we know what the control plane is actually doing, which matters when an overlay misbehaves and the vendor documentation stops being useful.
What’s included
- OSPF and BGP routing protocol design and implementation
- Network segmentation with VLANs and VXLAN overlays
- Firewall configuration with iptables/nftables or pfSense
- DNS server setup with BIND or CoreDNS
- DHCP service configuration and management
- Reverse proxy setup with Nginx or HAProxy
How we work
Addressing plans, routing policy, and segmentation boundaries get written down before anything is configured. The result should be a network your team can reason about from the documentation alone, without reverse-engineering a running device.