Network Fabric

Redundancy at every hop

Each host has multiple NIC paths into redundant switching fabrics, and upstream capacity comes from multiple tier-1 providers. The design goal is that no single failure — a NIC, a cable, a switch, a transit provider — is visible to a running workload. We alert on degraded redundancy rather than waiting for an outage, because losing one of two paths is worth acting on while everything still appears healthy: see status and monitoring.

Low latency across the Atlanta metro

Our Alpharetta, GA facility delivers sub-5ms latency across the Atlanta metro. For latency-sensitive work — synchronous replication, interactive applications, AI inference where tail latency decides whether the product feels responsive — physical proximity is not a detail you can optimise around later. Zone locations are listed under our zones.

Inter-zone traffic stays off the public internet

Traffic between our zones rides tier-1 backbones rather than the open internet and is encrypted in transit. That is what makes cross-zone replication and snapshot copies practical rather than merely possible — and because we do not charge egress, moving data between zones does not carry a per-gigabyte penalty. See storage architecture and pricing.

Where your traffic enters

Customer workloads sit in private VPCs reached over WireGuard, with load balancing in front of instance pools — described under networking and VPN. Where a shared internet path is not acceptable, we design dedicated circuits and BGP peering instead: see private circuits and peering.

Connecting hardware you own

Equipment racked with us under colocation connects into the same fabric as our compute platform, which is what allows one VPC to span your hardware and our cloud — the model described under hybrid cloud.

Have a specific latency or throughput requirement? Talk to an engineer.