AWS โ†” GCP & Service Selection

Coming from AWS? These are mappings, not equivalences โ€” the "key difference" column is where the architecture, delivery semantics, or pricing model actually diverge. Below, interactive matrices help you pick the right GCP service for the access pattern.

AWS โ†’ Google Cloud service map

Do not assume services are identical. Read the difference before you design around the analogy.

CategoryAWSGoogle CloudKey difference to remember
Compute (VM) EC2 Compute Engine Per-second billing, live migration, and sustained-use discounts are automatic on GCP; instance families and networking differ.
Functions Lambda Cloud Run functions Cloud Run functions run containers under the hood; Cloud Run blurs the FaaS/container line. Concurrency > 1 per instance is native on Cloud Run.
Serverless containers ECS Fargate / App Runner Cloud Run Cloud Run offers request-based billing with scale-to-zero and per-instance concurrency; no task definitions or cluster concepts.
Kubernetes EKS GKE GKE Autopilot removes node management entirely; GKE pioneered many upstream features. Control-plane pricing and node ops differ.
Object storage S3 Cloud Storage Single global API with regional/dual/multi-region locations; strong consistency. Storage classes and retrieval semantics differ from S3 tiers.
Relational DB RDS Cloud SQL Cloud SQL Auth Proxy / connectors and IAM DB auth differ from RDS security-group model.
Distributed Postgres Aurora PostgreSQL AlloyDB AlloyDB adds a columnar engine and AlloyDB AI (vectors) with full Postgres compatibility; architecture and pricing differ from Aurora.
Document/NoSQL DynamoDB Firestore Firestore is document-model with real-time listeners and offline SDKs; querying/indexing model differs sharply from DynamoDB's key-based design.
Wide-column DynamoDB (large scale) Bigtable Bigtable is a wide-column store for high-throughput/time-series; no secondary indexes, row-key design is everything.
Cache ElastiCache Memorystore Managed Redis/Memcached; HA, failover, and networking (private IP) models differ.
Queue SQS Cloud Tasks Cloud Tasks targets HTTP endpoints with per-task scheduling & rate limits; it's not a pull queue like SQS.
Pub/Sub messaging SNS + SQS Pub/Sub One service does fan-out + durable delivery, push & pull, ordering, and dead-letter โ€” different from SNS+SQS composition.
Event bus EventBridge Eventarc Eventarc uses CloudEvents and routes to Cloud Run/Workflows; audit-log-based event sources differ from EventBridge rules.
Orchestration Step Functions Workflows Workflows uses a YAML/JSON DSL with connectors; pricing is per-step. Composer (Airflow) covers data-pipeline orchestration.
Observability CloudWatch / X-Ray Cloud Operations (Logging/Monitoring/Trace) Integrated suite with Managed Service for Prometheus; log-based metrics and trace correlation differ.
Identity IAM roles IAM + service accounts GCP separates identities (service accounts) from roles; Workload Identity Federation replaces long-lived keys for external workloads.
GenAI platform Bedrock Vertex AI Vertex AI unifies managed models (Gemini), embeddings, vector search, tuning, and eval; SDK, grounding, and safety controls differ.
Vector search OpenSearch vector Vertex AI Vector Search Purpose-built ANN service; AlloyDB AI/pgvector is the SQL alternative. Indexing, filtering, and pricing differ from OpenSearch.
IaC AWS CDK / CloudFormation Terraform (Google provider) Terraform is the de-facto standard on GCP; Infrastructure Manager runs it managed. Config Connector manages resources via Kubernetes.

Service selection matrices

Match the access pattern โ€” data model, consistency, scale, latency, and cost โ€” to the service.

Database Selection

Compare across dimensions โ†’
Dimension Cloud SQLAlloyDBSpannerFirestoreBigtableBigQueryMemorystore
Data model RelationalRelational (+columnar/vector)Relational (distributed)DocumentWide-columnColumnar analyticsKey-value cache
Consistency StrongStrongStrong (global)StrongStrong (row)N/A (analytics)N/A (cache)
Horizontal scale Limited (replicas)High (read pools)UnlimitedHighVery highVery highHigh
Best latency msms (columnar for analytics)ms (global)mssingle-digit msseconds (queries)sub-ms
Vector search pgvector (extension)AlloyDB AI (native)NoLimitedNoYes (SQL)Semantic cache only
Best for OLTP appsOLTP + vectors + analyticsGlobal OLTP at scaleApp/mobile docsTime-series/IoTAnalytics warehouseCaching/sessions/rate-limit

Orchestration & Compute for Work

Compare across dimensions โ†’
Dimension WorkflowsCloud ComposerCloud Run JobsCloud TasksPub/SubBatchDataflowVertex AI Pipelines
Best for Service orchestration / sagasData-pipeline DAGs (Airflow)Finite container jobsHTTP task fan-outEvent fan-out/decouplingHPC/batch computeStream/batch data processingML pipelines
Duration Up to 1 yearLong (scheduled)HoursShort tasksN/A (messaging)LongLongLong
State Built-inAirflow metadata DBStateless taskPer-taskNonePer-jobManagedManaged
Human approval Yes (callbacks)PossibleNoNoNoNoNoPossible
Cost model Per stepEnvironment (always-on)Per job vCPU/memPer operationPer messagePer vCPU/memPer workerPer pipeline run

Storage Selection

Compare across dimensions โ†’
Dimension Cloud StoragePersistent DiskHyperdiskLocal SSDFilestoreBigQuery storage
Type ObjectBlockBlock (tunable)Block (ephemeral)File (NFS)Analytics columnar
Shared access Global (API)Single VM (or RO multi)Single VMSingle VMMany VMs (NFS)Query engine
Persistence DurableDurableDurableEphemeralDurableDurable
Best for Files, blobs, RAG docs, staticVM boot/data disksHigh-IOPS DB disksScratch/cacheShared POSIX filesystemsWarehouse tables