Microservices Monitoring & Load Testing Framework 🌩️📈

📌 Overview

This cornerstone project provides a high-fidelity benchmarking framework for hosting, monitoring, and stress-testing a microservices-based e-commerce application across diverse cloud topologies. The system spans Public (Azure), Private (IITD Baadal), Multi-Cloud (Azure+GCP), and Edge (K3s) environments, utilizing Kubernetes and a production-grade observability stack to evaluate performance trade-offs.

🏗️ Architecture

The project utilizes the Google Online Boutique (11-tier architecture) to simulate real-world traffic across four distinct infrastructures:

• Private Cloud: Baadal Cloud VMs (IIT Delhi).
• Public Cloud: Azure Multi-Node clusters.
• Multi-Cloud: Unified mesh across Azure and Google Cloud Platform (GCP).
• Edge Computing: K3s topology for frontend high availability.
• Networking: Tailscale VPN mesh for secure cross-cluster communication.

🚀 Key Features

• Multi-Cloud Orchestration: Deployed an 11-tier microservice architecture across heterogeneous cloud providers, managing pod distribution and cluster networking.
• Advanced Networking: Engineered a Tailscale VPN mesh to handle cross-cluster K8s routing, successfully bypassing institutional firewalls for private cloud integration.
• Edge High Availability: Implemented a K3s topology using nodeSelectors to isolate frontend replicas on dedicated edge nodes, optimizing for low-latency user access.
• Quantitative Observability:
  • Prometheus & Grafana: Correlated real-time CPU/Memory telemetry with application-level performance using PromQL.
  • Latency Benchmarking: Identified a 76% latency reduction in multi-node public cloud scaling vs. single-node setups.
• Resilience Analysis: Identified hardware ceilings in private infrastructure, documenting a critical saturation point at 3,530 ms latency under high load.

🛠️ Tech Stack

• Cloud: Microsoft Azure, Google Cloud Platform (GCP), IITD Baadal.
• Orchestration: Kubernetes (Kubeadm), K3s (Edge K8s).
• Networking: Tailscale (VPN Mesh), Flannel (CNI), Cloudflare Tunnel.
• Monitoring: Prometheus, Grafana, Node Exporter.
• Load Testing: Locust (Distributed Load Injection).

📊 Experimental Results (at 200 Users)

Environment	Avg Latency	Throughput (RPS)	CPU Saturation	Memory Saturation
Azure (Public)	148 ms	61.65	14.7%	43.1%
Baadal (Private)	3,530 ms	29.57	69.5%	90.8%
Multi-Cloud	3,633 ms	34.1	45.5%	69.9%

---

🗺️ Project Roadmap

Phase 1: Private Cloud Baseline ✅ Completed

• Established a fully observable, production-grade microservices environment on Baadal Cloud.
• Engineered Split-Horizon proxy bypass and Cloudflare Tunnels for external access.

Phase 2: Public Cloud Implementation ✅ Completed

• Provisioned multi-node Azure clusters and replicated the observability stack.
• Benchmarked hardware utilization vs. private cloud ceilings.

Phase 3: Hybrid Cloud Engineering (Pivoted) ✅ Completed

• Attempted private-to-public NAT traversal; pivoted to Multi-Cloud VPN mesh due to infrastructure constraints.

Phase 4: Multi-Cloud VPN Mesh (Tailscale) ✅ Completed

• Engineered a Tailscale VPN mesh for cross-cluster K8s routing across Azure and GCP.
• Documented the ~10x network overhead tradeoff introduced by the encrypted tunnel.

Phase 5: Aggregated Load Testing & Analytics ✅ Completed

• Executed tiered Locust simulations (10, 50, 200 users) and analyzed performance-to-saturation correlations.

Phase 6: Edge Computing Extension (K3s) ✅ Completed

• Architected a K3s topology to isolate frontend high-availability workloads at the edge node.

---

💡 Future Work

• AI-Driven Traffic Routing: Implementing workload orchestration that dynamically shifts traffic based on real-time latency and cloud spot-instance pricing.
• Chaos Engineering: Integrating tools to simulate network partitions and pod failures across the multi-cloud mesh to test system self-healing.
• Deep IoT Integration: Extending edge nodes to support real-world sensor data ingestion and processing.