Carrier-grade GCP Core
Pioneered PCF and voice-core deployment on Google Cloud for a major European telecom environment — building the GKE foundation, solving networking below the Kubernetes abstraction, and validating carrier-grade workloads in production.
Challenge
A cloud-native telecom core had to run on Google Cloud for the first time. The workload required GRE-based signalling between network functions spread across GKE nodes — but the managed cluster dataplane did not permit the traffic path the application depended on. Standard Kubernetes networking was not enough, and nothing about the failure was visible from the application layer.
Investigation
The failure looked like a simple connectivity problem and was anything but. I worked down the stack layer by layer, ruling each one in or out with evidence rather than assumption.
- Traced packet paths hop by hop rather than trusting service-level reachability tests.
- Examined the managed cluster's eBPF-based dataplane behaviour for the specific protocol in question.
- Confirmed cloud firewall rules permitted the protocol — ruling out the most commonly blamed layer.
- Investigated cross-node reachability, pod placement, routing tables, IPv6 addressing and neighbour discovery on secondary interfaces.
- Separated L3/routing/firewall causes from L2 neighbour-resolution behaviour using ip route, ip addr, ip neigh and tcpdump.
Solution
Rather than fight the managed dataplane, I gave the signalling traffic its own path: a secondary pod interface dedicated to signalling, leaving default cluster networking untouched for everything else.
My contribution
- Built the GCP foundation: GKE clusters, VPCs, subnets, node pools, static IPs, Cloud DNS, TLS, ingress, IAM and Workload Identity.
- Owned the networking investigation end to end, producing evidence-based findings that separated application, Kubernetes, host and cloud-provider responsibilities.
- Designed and validated the secondary-interface signalling path so the core could operate without compromising default cluster networking.
- Documented symptoms, commands, confirmed non-issues and next steps for both internal engineering and cloud-provider escalation.
- Managed Crossplane-driven GCP resources, cost controls and environment lifecycle across validation and demonstration environments.
Impact
The core ran on Google Cloud in a carrier-grade configuration, unblocking a customer engagement that had no precedent to follow. The networking pattern and the troubleshooting method became reusable for later deployments.