DevOps for Telecom & Connectivity platforms

What regulates the infrastructure

What actually goes wrong here

• Signalling storms: after a cell, core element or software upgrade restarts, all affected subscribers reattach at once and overwhelm the HSS/HLR, IMS registration or AMF, so recovery from a small fault triggers a much larger outage than the fault itself
• Provisioning and OSS/BSS failure, where the radio network is entirely healthy but new activations, SIM swaps, top-ups and prepaid balance checks stop — commercially severe and largely invisible to network-level monitoring
• Emergency call routing or caller-location failure, which is a regulatory incident with mandatory notification and potential enforcement regardless of how few calls were affected
• Submarine cable cuts, where the operational problem is not the loss itself but repair time: cable ships must be contracted and given permits, so repairs in the Red Sea corridor during 2024-2025 ran from about three weeks to five or six months — the February 2024 event severed several systems at once and AAE-1, Seacom and EIG were not restored until mid-to-late 2024 — forcing sustained operation on restoration capacity with degraded latency
• Terrestrial fibre cuts from civil works and road construction, the single most common cause of backhaul loss, whose impact depends entirely on whether the affected route is genuinely diverse or only nominally so
• BGP misconfiguration, route leaks and hijacks propagating an operator's error outward to peers and transit customers
• Signalling-protocol abuse and fraud over SS7, Diameter and SIP — international revenue share fraud, Wangiri, SIM-box bypass and location or SMS interception — which is a direct revenue loss rather than an availability event
• Cell site power failure during grid outages, where the practical limit is battery autonomy and refuelling logistics across thousands of distributed sites, not equipment reliability
• GNSS or PTP timing loss, which desynchronises radio units and degrades or drops service without any component actually failing
• DNS or subscriber-facing platform failure making the network appear down to customers while transport is fully operational

How demand behaves

Continuous 24/7 with a pronounced evening busy hour, and networks are dimensioned on that busy hour rather than on average load. Two load dimensions move independently: bearer traffic, dominated by video streaming and driven by content events, and signalling load, which scales with the number of attached devices and their mobility rather than with bandwidth — so a network can be far from its throughput limit while its signalling plane is saturated. Mass gatherings concentrate both into single cells, and emergencies produce simultaneous nationwide spikes.

Data you will be holding

Operators hold traffic and location metadata, which under the ePrivacy Directive 2002/58/EC is subject to confidentiality and erasure or anonymisation duties distinct from ordinary GDPR personal data. National data retention regimes vary and have been repeatedly narrowed by CJEU case law. Operators are also legally obliged to maintain lawful interception capability, which means production networks carry a mandated, tightly access-controlled interception interface — infrastructure that exists purely for legal reasons and whose compromise is itself a reportable incident.

Architecture this pushes you toward

Telecom infrastructure separates into domains with very different characteristics: RAN, packet core, IMS, transport, and OSS/BSS, with the last resembling ordinary enterprise IT and frequently being the weakest link. The 5G standalone core is cloud-native — containerised network functions on Kubernetes — which has moved traditionally bespoke telco workloads onto general-purpose compute while retaining hard requirements that general-purpose platforms do not usually meet: sub-millisecond determinism, precise clock synchronisation, and geographic N+1 redundancy of core sites. Edge compute sites are constrained by physical estate, power and access rather than by capacity planning.

Availability expectation

Core network elements are engineered to carrier-grade norms, conventionally expressed as 99.999% (about five minutes per year), and this expectation is structural rather than merely contractual: licence conditions typically impose coverage and quality-of-service obligations, and emergency call access carries availability duties in its own right. Enterprise connectivity products are sold with lower, explicitly contracted SLAs backed by service credits.