Timeline & Info
| Start date | End date | Associated with | Resources |
|---|---|---|---|
| April 2024 | September 2024 | Faculty of Informatics and Digital Technologies, University of Rijeka | Thesis PDF · GitHub · Defense Video |
Overview
Deployed and benchmarked Google’s “Online Boutique” (15-microservice e-commerce application) on both cloud-based (AKS) and on-premises (K3S) infrastructure with identical resources. Conducted comprehensive performance testing, cost analysis, and qualitative comparison to develop a selection framework for organizations choosing Kubernetes platforms.
Key Findings
Performance Results
K3S outperforms AKS by 3-5x in raw performance:
| Metric | K3S | AKS | Winner |
|---|---|---|---|
| Throughput | 109.44 req/s | 34 req/s | K3S (3.2x) |
| Response Time | 9.14ms avg | 29.41ms avg | K3S |
| CPU Performance | 20,219 events/s | 8,051 events/s | K3S (2.5x) |
| Idle CPU Usage | 12.5% | 42.3% | K3S (3.4x less) |
| Failed Requests (500 users) | <0.1% | 2.96% | K3S |
Cost Analysis (5-Year TCO)
AKS is 2.25x more cost-effective:
| Platform | Total Cost | Main Components |
|---|---|---|
| AKS | $12,499 | Compute: $8,760 · Storage: $730 · Networking: $1,825 · Labor: $1,184 |
| K3S | $28,135 | Hardware: $10,000 · Power: $4,380 · Maintenance: $2,470 · Labor: $8,000 |
Key Insight: AKS wins on cost unless you have existing on-premises infrastructure with high utilization.
Quick Comparison
| Aspect | AKS | K3S |
|---|---|---|
| Performance | Good | Excellent (3x faster) |
| Cost (5yr) | $12.5K | $28K |
| Setup Time | 15-30 min | 45-90 min |
| Management | Automated | Manual |
| Customization | Limited | Full control |
| Best For | Cloud-first, ease of use | Performance, on-prem, edge |
Tech Stack
Platforms
- Azure Kubernetes Service (AKS) - Managed Kubernetes on Azure
- K3S - Lightweight Kubernetes by Rancher
Infrastructure
- Azure (AKS): Terraform IaC, Standard_A2_v2 VMs (2 vCPU, 4GB RAM)
- VMware vSphere (K3S): Ansible automation, Rocky Linux, 3 control + 2 worker nodes
Application
- Google Online Boutique - 15-microservice e-commerce demo
- Languages: Go, Java, Node.js, Python, C#
- Communication: gRPC, HTTP/REST
- Storage: Redis
Testing & Monitoring
- Load Testing: Apache Benchmark, K6 (up to 500 concurrent users)
- Benchmarking: Sysbench (CPU, memory, I/O)
- Monitoring: Prometheus, Grafana, kube-prometheus-stack
Automation
- Terraform - Azure infrastructure provisioning
- Ansible - K3S cluster automation
- Helm - Kubernetes package management
- kubectl - Cluster management
Architecture
AKS Deployment
- Azure Resource Group → AKS Cluster → VM Scale Sets → Azure Load Balancer
- Networking: Kubenet + Calico network policies
- Fully automated via Terraform
K3S Deployment
- VMware vSphere → HA Control Plane (3 nodes) → Worker Nodes (2)
- Networking: Flannel CNI
- Automated via Ansible playbooks
Online Boutique App
15 microservices: Frontend, Product Catalog, Cart, Checkout, Payment, Shipping, Currency, Recommendation, Email, Ad Service, Load Generator
Recommendations
Choose AKS When:
- Cost is primary concern (no existing infrastructure)
- Ease of management > raw performance
- Azure ecosystem integration valuable
- Limited Kubernetes expertise in-house
- Cloud-first strategy
Choose K3S When:
- Maximum performance critical
- Existing on-premises infrastructure
- Edge/IoT deployments
- Need full customization
- Data sovereignty requirements
- Expert Kubernetes team available
Skills Demonstrated
Cloud & Infrastructure: Azure, Kubernetes, VMware vSphere, Infrastructure as Code
Automation: Terraform, Ansible, CI/CD concepts
Containers: Docker, Kubernetes orchestration
Monitoring: Prometheus, Grafana, performance analysis
Testing: Load testing, benchmarking, system optimization
Networking: Kubernetes networking, load balancing, CNI plugins
Research: Quantitative analysis, cost modeling, technical writing
Supported by: SICK Mobilisis (infrastructure resources)