# Building a Community Mesh-Network ISP That Can Survive Bad Weather

Our neighborhood mesh ISP started as three rooftops, two donated sector antennas, and a spreadsheet of people who could climb ladders safely. Two years later, it serves 140 households across a mix of row houses, apartments, and small businesses. The goal was never to beat fiber on peak throughput; it was to provide a locally operated network with transparent pricing, resilient local paths, and enough upstream capacity that remote work and school did not collapse during outages.

## Network Design

Each roof node has at least two roles: access for nearby subscribers and backhaul to other roofs. We use 5 GHz point-to-point links for most backhaul, 60 GHz where line-of-sight is excellent, and 2.4 GHz only for short access links where foliage makes higher bands unreliable. The mesh is not a flat Wi-Fi cloud; every node routes, every link is measured, and dense areas get multiple paths so a single failed radio or wet tree does not isolate a block.

## Routing

The routing layer uses Babel because it handles lossy wireless links well and reacts quickly without needing a central controller. Each node advertises its local subscriber subnet, while border routers inject a default route only when their upstream health checks pass. A typical edge node configuration is intentionally small:

```ini
# /etc/babeld.conf
interface mesh0 type wireless
interface eth0.20 type wired
redistribute local ip 10.42.18.0/24
redistribute local deny
```

## Operations

The hardest problems have been operational, not theoretical. We track radio alignment, signal-to-noise ratio, airtime utilization, and packet loss from a small Prometheus stack, but the most useful alert is still “this link has been getting worse for seven days.” Slow degradation usually means a shifted mount, new construction, water in a connector, or summer leaves growing into the Fresnel zone.

## Lessons Learned

A community ISP works when the network design matches the social model. We publish node maps, maintenance windows, and postmortems; subscribers can see why an outage happened and what changed afterward. That transparency does not make wireless backhaul immune to weather or congestion, but it does turn the network into shared infrastructure instead of a black box owned by someone far away.