Cloud-native architecture on GCP and AWS
Networking, Kubernetes, platform controls and delivery pipelines for enterprise workloads. By Sanjeev Kumar, Application Architect.
GCP-first, AWS-fluent
My primary cloud in recent delivery is Google Cloud Platform, with AWS as a strong second — reflected across the portfolio (GCP networking / VPC Service Controls / Interconnect, BigQuery lakes, and dual-cloud field-service platforms). Cloud choice follows data residency, existing estates and team skill; architecture principles stay portable.
Network and perimeter
Enterprise programmes live or die on networking: shared VPCs, private service connect / PrivateLink, interconnects to on-prem, and perimeter controls that keep regulated data from drifting to the public internet. Identity-aware access and least-privilege service accounts beat long-lived keys in CI. Terraform (or equivalent IaC) makes those topologies reviewable.
Kubernetes and runtime
Containers on GKE/EKS need resource requests that match reality, pod disruption budgets, ingress policy and secret injection that does not bake credentials into images. Health checks must mean “ready to take traffic”, not merely “JVM process is up”. Deeper notes for enterprise Java on Kubernetes live on the Kubernetes architecture page; the runtime diagram shows the ingress → service → pod shape.
Data platforms and pipelines
Analytics paths (for example BigQuery-backed repositories on Equifax DMP and GCP platform work in the portfolio) need clear producers, schema ownership and quality checks. Streaming joins Kafka or event pipelines (or Pub/Sub on field-service estates) when operational events must land near-real-time. Private endpoints and VPC controls also matter when enterprise AI workloads call hosted model APIs.