AWS — Services & Pricing

作者：Sajiv Francis 约 3 分钟阅读更新日期：2026-07-22 标签：云架构, Cloud Architecture, 架构, 多云

Amazon Web Services is the broadest and most mature of the three majors — the widest service catalog, the deepest set of primitives, and the largest partner/marketplace ecosystem. Its design center is composable building blocks: you assemble fine-grained services (S3, EC2, Lambda, SQS, IAM) rather than adopt opinionated end-to-end stacks. That breadth is its strength and its tax — maximum flexibility, but more pieces to wire and govern.

This page localizes the reference architecture to AWS: CloudFront → API Gateway / ALB → EC2 · EKS · Lambda → RDS / DynamoDB / ElastiCache / S3 → Redshift · Kinesis · SageMaker, with IAM, KMS, CloudWatch, and CloudFormation cross-cutting.

Services by capability

Capability	Service	What it is	Unit price	Source
Object Storage	S3 Standard	Durable object storage for unstructured data; standard tier for frequently accessed objects.	$0.023/GB-month	live
Block Storage (SSD volume)	EBS gp3	General-purpose SSD block volume for EC2; baseline IOPS/throughput included.	$0.08/GB-month	live
VM Compute (general purpose, ~2 vCPU / 8 GB)	EC2 m5.large (2 vCPU)	General-purpose on-demand VM, 2 vCPU / 8 GB; price normalized per vCPU-hour.	$0.048/vCPU-hour	live
Serverless Functions / Edge Compute	Lambda	Event-driven functions; first 1M requests/mo free, then per-request.	$0.2/1M requests	list
Managed Relational DB (PostgreSQL, ~2 vCPU / 8 GB)	RDS PostgreSQL db.m5.large	Managed PostgreSQL, 2 vCPU / 8 GB, single-AZ on-demand (instance only).	$0.178/instance-hour	list
Internet Data Egress	EC2/S3 egress	Internet data transfer out; first 100 GB/mo free, then per-GB.	$0.09/GB	list
CDN (content delivery egress)	CloudFront	CDN cache egress (NA/EU first tier).	$0.085/GB	list
Managed Kubernetes (control plane)	EKS	Managed Kubernetes control plane (per-cluster fee; nodes billed separately).	$0.1/cluster-hour	list

Core building blocks

Compute — EC2 for VMs (deepest instance-family selection of any cloud), Lambda for event-driven serverless, ECS/Fargate for containers without cluster ops, EKS for managed Kubernetes. Spot and Savings Plans are the primary cost levers.
Storage — S3 is the de-facto object-storage standard (durability, lifecycle tiering, event triggers); EBS for block volumes, EFS/FSx for shared file systems.
Data — RDS/Aurora (Aurora is AWS’s high-performance MySQL/PostgreSQL-compatible engine), DynamoDB for serverless NoSQL at scale, Redshift for warehousing, ElastiCache for Redis.
Integration & events — SQS/SNS/EventBridge and Step Functions for decoupled, event-driven architectures — an area where AWS’s primitives are especially rich.
AI/ML — SageMaker for the full ML lifecycle; Bedrock for managed foundation models.
Governance — IAM (the most granular permission model of the three), KMS, CloudWatch, CloudFormation/CDK for infrastructure-as-code.

When to choose AWS

You want maximum breadth and the richest primitives — niche services, the most instance types, the deepest event/integration toolkit.
Largest talent pool and ecosystem — most third-party tools, SDKs, and certified engineers target AWS first.
Serverless-first or event-driven designs — Lambda + EventBridge + Step Functions is a mature, well-trodden pattern.
Lean toward elsewhere when: you want lower-friction, more opinionated managed stacks (GCP’s data/AI; Azure’s enterprise-Microsoft integration), a simpler bill, or you’re optimizing egress-heavy workloads (see Cloudflare’s $0 egress). AWS’s flexibility means more services to configure and a more complex cost surface.

Cost note: the table above is the first-tier on-demand list/live price per capability. AWS’s real-world economics shift substantially with Savings Plans / Reserved Instances (compute) and S3 tiering — model those for committed workloads.

Pricing generated from the live cost catalog via scripts/gen_cloud_docs.py.

来源说明：本文翻译/整理自 docs.sajivfrancis.com，作者 Sajiv Francis。原文链接保留，版权归原作者所有。