Design. Deploy. Observe.
Experiment. Destroy.
A personal laboratory for comparative cloud systems engineering across AWS, GCP, and Azure. Building evidence-driven reference architectures, disposable cloud environments, and automated lifecycle guardrails.
Core Cloud Providers & Architecture Ecosystem
Comparative engineering across leading enterprise cloud platforms
Eight core principles guiding every cloud experiment.
The foundational engineering rules defined in the SymmetricCloud Architecture Vision. Designed for deep learning, empirical verification, and bulletproof cost safety.
Learning Before Complexity
No infrastructure is introduced merely because it is architecturally fashionable. Every component exists to answer a specific architectural hypothesis.
Disposable by Default
Experimental resources are strictly temporary unless explicitly designated permanent. Automated TTL policies destroy workloads after tests conclude.
Automation Over Manual Repetition
Every environment is defined in declarative Terraform. Deployments, assertions, telemetry captures, and teardowns are fully automated.
Evidence Over Claims
Architectural assertions require empirical proof: measured p95 latencies, failure injection logs, telemetry traces, and actual cloud cost attribution.
Provider Nuances Are Valued
AWS, GCP, and Azure are evaluated for their native strengths, managed service differences, and operational limits rather than forced into artificial parity.
Secure by Default & Zero Keys
The public UI never touches cloud credentials. Deploy and destroy operations require authentication with short-lived OIDC IAM session roles.
Strict Cost Guardrails
Hard budget alarms and per-experiment limits protect against runaway cloud spend. Zero idle resources when experiments are inactive.
Minimal Control Plane
The management system remains simpler than the architectures it manages. Python FastAPI orchestration and clean job queues.
Six core architecture domains, rigorously investigated.
Every track produces reproducible Terraform modules, architectural decision records (ADRs), and live experiment evidence across cloud providers.
The 7-stage loop: Design to Destroy.
Every cloud architecture investigation adheres to an evidence-driven lifecycle. Workloads are provisioned with strict TTL policies, measured under load, documented for insights, and cleanly torn down.
- 1. Design & Hypothesis
Architectural questions are formalized before writing code. Multi-cloud tradeoffs are defined in ADRs.
- 2. Declarative IaC Deploy
Every environment is defined in modular Terraform with mandatory TTL tags and budget constraints.
- 3. Observe & Measure
Empirical verification using OpenTelemetry, CloudWatch, and real latency/throughput measurements.
- 4. Verified Destruction
Experiments are cleanly torn down. Zero lingering cloud resources or runaway monthly cloud bills.
# experiment-workload.tf
module "serverless_lab" {
source = "./modules/aws-serverless-api"
project = "symmetriccloud"
environment = "ephemeral"
ttl_hours = 4 # Auto-destroy guardrail
runtime = "python3.12"
memory_size = 1024
concurrency = 100
tags = {
Owner = "ArunKumar"
Lifecycle = "Disposable"
Purpose = "Experiment-401"
}
}A single pane of glass over your global footprint.
Watch every cluster, region and dependency in real time. Symmetric continuously reconciles desired state against what's actually running — across clouds — and surfaces drift, cost and latency in one view.
Strict security boundaries. Zero runaway cloud costs.
How the SymmetricCloud laboratory protects credentials, restricts privileged Terraform actions behind authenticated login, and guarantees disposable infrastructure.
Only the authenticated laboratory owner can trigger infrastructure provisioning or destruction. Zero public execution endpoints or open webhooks.
auth-gated controlsSource code, Terraform states, and sensitive experiment modules are hosted in private GitHub repositories with push protection enabled.
sealed source codeNo long-lived AWS_SECRET_ACCESS_KEYs stored in code or client storage. Short-lived IAM session tokens are generated via OIDC federation.
oidc federated iamEvery provisioned test stack carries a Time-To-Live countdown tag (typically 2 to 4 hours). Automated cleanup scripts destroy idle resources.
ephemeral automationAWS Budgets and billing alarm guardrails automatically trigger SNS notifications and emergency teardown if spend exceeds defined thresholds.
finops hard capsCore portfolio assets run on S3, CloudFront, and Lambda on-demand primitives — guaranteeing $0.00 idle cost when no experiments are running.
$0 idle baselineHigh-level case study briefs, architecture diagrams, and cloud comparison tables. Zero cloud keys or privileged controls.
- Private GitHub repositories & Terraform codebases
- Deploy & Destroy operational actions with TTL limits
- Live CloudWatch & OpenTelemetry experiment traces
Measured lab telemetry, not marketing claims.
Observed performance data, teardown success rates, and cost telemetry captured across experimental workloads.
Empirical reference architectures, with public live demos.
Public summaries are open for every case study. Disposable workloads reflect real-time lifecycle states (RUNNING vs STOPPED) so standing costs remain $0.00 when dormant, while deep implementation code and privileged control planes remain securely authenticated.
A complete serverless web application featuring DynamoDB single-table design, Lambda event authoring, REST API Gateway, and CloudFront static hosting.
- frontend
- React 19 + Vite
- backend
- Node.js 20 Lambda
- database
- DynamoDB (on-demand)
Production-grade e-commerce marketplace investigating multi-tenant PostgreSQL schema partitioning, FastAPI async endpoints, Okta auth, and Stripe billing.
- frontend
- React 18 + Vite + Tailwind
- backend
- FastAPI + SQLAlchemy
- database
- PostgreSQL + MongoDB
Embedded e-commerce partner application with OAuth 2.0 handshake, HMAC signature verification, order webhooks, and Stripe subscriptions on AWS Lambda.
- frontend
- React 18 + Vite + Tailwind
- backend
- FastAPI + Mangum (Lambda)
- database
- DynamoDB (4 tables)
High-performance media transcoding pipeline utilizing Python 3.12, PyAV native bindings, and direct S3 presigned URL streaming.
- frontend
- Vanilla JS Drag-and-Drop + S3 Presigned Upload
- backend
- Python 3.12 Lambda + PyAV Transcoder
- storage
- S3 Presigned Direct Upload & Download
Comparative event ingestion evaluating AWS EventBridge/SQS against GCP Pub/Sub for resilient message fan-out and dead-letter queue handling.
- frontend
- React Observability Dashboard (Private)
- backend
- Python Event Processors
- database
- DynamoDB Streams
Resilient microservices mesh featuring service discovery, distributed tracing via OpenTelemetry, and circuit breaker patterns on Kubernetes.
- frontend
- Angular 17 Control Panel (Private)
- backend
- Spring Boot 3.x Reactive
- database
- PostgreSQL + Redis Cluster
Modern enterprise storefront evaluating Angular Server-Side Rendering (SSR), Express.js backend, and Azure Cosmos DB multi-region latency.
- frontend
- Angular 17 SSR
- backend
- Node.js Express + Azure Functions
- database
- Azure Cosmos DB (Serverless)
Comparative runtime benchmarking of serverless API endpoints implemented in Java 21, Python 3.12, Rust, and Node.js 20 with shared API Gateway.
- frontend
- Next.js Benchmark Visualizer (Private)
- backend
- Lambda (Rust + Java + Python + Node)
- database
- DynamoDB + Aurora Serverless
Real-time Workload Lifecycle & Zero Idle Cost Policy
Ephemeral Lifecycle: Demos are scaled to zero ($0.00 idle cost) when inactive. When launched, automated TTL sweeps ensure resources terminate cleanly after evaluation.
Privileged Control Plane: 1-click deploy/destroy actions, raw Terraform state, and secrets vault are restricted to authenticated owners and guest auditors.
AWS vs. GCP vs. Azure: Architectural Mapping
Comparative analysis of cloud primitives across the leading providers. Rather than forcing artificial parity, SymmetricCloud investigates the native strengths and operational tradeoffs of each provider.
| Architecture Primitive | Amazon Web Services | Google Cloud Platform | Microsoft Azure |
|---|---|---|---|
Serverless FunctionsAWS leads in cold-start tuning; GCP offers unified container packaging. | AWS Lambda (x86_64 & Graviton) | Cloud Functions (2nd Gen / Cloud Run) | Azure Functions (Flex Consumption) |
Container WorkloadsCloud Run is the fastest to deploy; EKS offers the most mature IAM integration. | ECS Fargate & EKS | Cloud Run & GKE Autopilot | Azure Container Apps & AKS |
Object Storage & CDNAWS CloudFront OAC provides gold-standard least-privilege bucket access. | S3 + CloudFront (OAC) | Cloud Storage + Cloud CDN | Blob Storage + Azure Front Door |
NoSQL DatabasesDynamoDB delivers sub-10ms predictability; Firestore excels in real-time listeners. | DynamoDB (Single-table, on-demand) | Firestore / Bigtable | Cosmos DB (Serverless) |
Identity & SecretsAll three support short-lived federated credentials with zero stored keys. | IAM OIDC + Secrets Manager | Workload Identity + Secret Manager | Managed Identities + Key Vault |
Design. Deploy. Observe.
Experiment. Destroy.
A personal engineering laboratory dedicated to comparing real cloud architectures, measuring resilience, and governing multi-cloud deployments with strict security and zero-standing credentials.