Blog / Technical / Cloud Architecture

Cloud Architecture

Cloud architecture leverages managed cloud services to build scalable, resilient, and cost-efficient systems without managing physical hardware. The cloud provider's shared responsibility model, global...

Shreedhar Kodate · 18 October 2025 · 3 min read
large-scale-software-engineeringdocs

Cloud architecture leverages managed cloud services to build scalable, resilient, and cost-efficient systems without managing physical hardware. The cloud provider's shared responsibility model, global infrastructure, and broad service catalog enable capabilities that were previously only available to large enterprises.

Cloud Service Models

graph TD subgraph IaaS[IaaS - Infrastructure as a Service] I1[You manage: OS Apps Data Runtime] I2[Provider manages: Virtualization Servers Storage Network] I3[Examples: EC2 GCE Azure VMs] style I3 fill:#fef3c7,stroke:#d97706 end subgraph PaaS[PaaS - Platform as a Service] P1[You manage: Applications Data] P2[Provider manages: Runtime OS Infra] P3[Examples: Heroku App Engine Cloud Run] style P3 fill:#dbeafe,stroke:#2563eb end subgraph SaaS[SaaS - Software as a Service] S1[You manage: Data and configuration] S2[Provider manages: Everything] S3[Examples: Salesforce Gmail Datadog] style S3 fill:#dcfce7,stroke:#16a34a end IaaS --> PaaS --> SaaS

AWS Well-Architected Framework

mindmap root((Well-Architected\nFramework)) Operational Excellence Infrastructure as code Frequent small reversible changes Anticipate failure Learn from operational events Security Strong identity foundation Enable traceability Apply security at all layers Protect data in transit and at rest Reliability Automatic recovery from failure Scale horizontally Stop guessing capacity Manage change with automation Performance Efficiency Democratize advanced technologies Go global in minutes Use serverless architectures Experiment more often Cost Optimization Adopt consumption model Measure overall efficiency Analyze and attribute expenditure Use managed services Sustainability Understand your impact Maximize utilization Use efficient hardware

Multi-Region Architecture

graph TD subgraph Global[Global Services] Route53[Route 53\nGlobal DNS\nHealth-based routing] CF[CloudFront CDN\nGlobal edge network\n300+ PoPs] IAM[IAM\nGlobal identity] end subgraph Primary[Primary Region: us-east-1] ALB1[Application Load Balancer] ECS1[ECS / EKS Cluster] RDS1[(RDS Primary\nMulti-AZ)] ALB1 --> ECS1 --> RDS1 end subgraph DR[DR Region: us-west-2] ALB2[Application Load Balancer] ECS2[ECS / EKS Cluster\nstale or cold] RDS2[(RDS Read Replica\nor Aurora Global)] ALB2 --> ECS2 --> RDS2 end Route53 -->|primary| ALB1 Route53 -->|failover on health check fail| ALB2 RDS1 -->|async replication| RDS2 CF --> ECS1 & ECS2 style Primary fill:#dcfce7,stroke:#16a34a style DR fill:#fef3c7,stroke:#d97706

Key Concepts

  • IaaS (Infrastructure as a Service): Cloud provider offers virtualized compute, storage, and networking. Customer manages OS, runtime, and applications. Maximum control, maximum management responsibility. EC2 is the canonical example.

  • PaaS (Platform as a Service): Cloud provider manages the runtime environment; customers deploy applications. Lower operational overhead — no OS patching, no runtime management. Less control. Examples: Google App Engine, AWS Elastic Beanstalk, Cloud Run.

  • Serverless: Cloud provider manages all infrastructure including scaling. Pay per invocation. Zero idle cost. Maximum operational simplicity for stateless event-driven workloads.

  • Shared Responsibility Model: The cloud provider secures the infrastructure (physical datacenters, hypervisors, global network). The customer secures what they deploy (OS configuration, network security groups, IAM policies, application code, data encryption).

  • Availability Zones (AZs): Physically separate datacenters within a region, connected by low-latency high-bandwidth links. Deploying across multiple AZs provides resilience against single-facility failures. Most cloud services offer Multi-AZ deployment as a standard option.

  • Cost Optimization: Cloud costs can grow unexpectedly. Strategies: right-size instances (don't over-provision), use Spot/Preemptible instances for fault-tolerant workloads (60-90% discount), use Reserved Instances for predictable base load (30-60% discount), implement lifecycle policies to delete old snapshots/logs, use Savings Plans for committed usage.

  • Cloud Native: Designing systems that exploit cloud capabilities — auto-scaling, managed services, pay-per-use, global distribution — rather than lifting-and-shifting on-premises architectures to the cloud.

Trade-offs

Approach Control Operational Cost Portability
IaaS (EC2) High High Medium
PaaS (Cloud Run) Medium Low Medium
Serverless Low Very Low Lower
On-premises Highest Highest Full

When to Use

  • IaaS: When you need specific hardware, custom OS configuration, or capabilities not available as managed services
  • PaaS: Stateless web applications, API backends — significant reduction in operational overhead
  • Serverless: Event-driven processing, variable/spiky workloads, rapid prototyping
  • Multi-region: When latency to a single region is unacceptable for global users, or when disaster recovery requirements mandate geographic redundancy