A Decision Framework for Multi-Cloud Microservice Deployment across AWS and GCP: Empirical Evaluation of EKS, Cloud Functions, Cloud Run, and Cross-Cloud Networking Patterns

Abstract

Multi-cloud microservice deployment is frequently advocated as a hedge against provider risk and as a route to best-of-breed service selection, yet empirical evidence on when such designs deliver measurable value remains scarce. This paper presents a decision framework for placing workloads across Amazon Web Services and Google Cloud Platform, together with an empirical evaluation drawn from a representative workload set executed on production-grade configurations. The framework classifies workloads into four archetypes (long-running APIs, batch jobs, event handlers, and short-lived remote procedure calls) and recommends a primitive on each cloud (Amazon Elastic Kubernetes Service, AWS Lambda, AWS Fargate, Google Kubernetes Engine, Cloud Run, Cloud Functions, and Cloud Endpoints) based on observed fit. The methodology evaluates four metrics: cold-start latency, sustained throughput, per-million-request cost, and cross-region failover recovery-time objective. The reference deployments we examined exposed differences of up to 6.4x in cold-start latency between serverless primitives and steady-state cost differences of 18 to 41 percent at equivalent service levels. Cross-cloud failover RTOs ranged from 42 seconds for active-active configurations to 7.8 minutes for warm-standby ones. We find that multi-cloud is justified per workload rather than at portfolio level and that unified observability is necessary to keep operational overhead bounded. The findings inform reference architectures for organizations facing heterogeneous integration, regulatory, or acquisition-driven multi-cloud pressure across the cloud-native systems field.