ADR 0001: Mobile core - Open5GS instead of Magma (or free5GC)¶

Context¶

Fairwave needs a 4G EPC and a future path to 5G GC. Candidates:

Open5GS - C, standalone binaries, tiny memory footprint, REST-ish configs, active community, 4G+5G SA in one repo.
Magma - Python/Go gateway and cloud-native design; heavier ops burden, more moving parts.
free5GC - pure Go; strong 5G focus, smaller 4G track record.
OAI CN - very complete but operationally complex, many knobs.

Use Open5GS as the default mobile core, driven by fairwave-control via configuration templating, Helm, and Compose.

[+] Small memory footprint - fits pizza-box targets (NUC, CM4/5).
[+] Single repo with MME+SGW+PGW+HSS+PCRF+SMF+UPF+UDM+AUSF; simpler mental model.
[+] AGPL-3.0 is acceptable for Fairwave's posture (we configure and drive; no source modification by default).
[+] Fast gNB/eNB interoperability via srsRAN containers.
[!] No built-in northbound REST for provisioning; we must implement it in fairwave-control.
[!] Upstream CVEs propagate; hence pinned digests, SBOM, cosign signing.

Magma - rejected: ops burden too large for a pizza-box, cloud-first model opposes local-first.
free5GC - deferred: could be the 5G SA path later via a feature flag; 4G remains Open5GS.
OAI CN - rejected: complexity too high for community operators; revisit later for RU/DU split research.