Cutting 44% Azure Spend for a US Logistics Platform using AKS & Spot Instances
A real-time logistics tracking platform handling 80M daily IoT telemetry events saw their Azure monthly burn skyrocket to $45,200/mo. We completely restructured their compute topology into Azure Kubernetes Service with KEDA spot auto-scaling.
The Challenge & Technical Inefficiencies
The client operated 28 static standard D16s_v4 virtual machines running 24/7 to absorb peak morning delivery hours. Over 60% of CPU cycles were sitting idle between 8 PM and 5 AM. Furthermore, 42 unattached Premium SSD managed disks from decommissioned nodes were quietly racking up $3,800/mo in zombie billing.
Our Engineering Solution
- • Containerization: Dockerized backend ingestion services and deployed on an Azure Kubernetes Service (AKS) cluster.
- • Spot Instance Integration: Shifted non-critical telemetry queue consumers to Azure Spot node pools at an 82% discount, using graceful termination webhooks.
- • KEDA Event-Driven Sizing: Configured Kubernetes Event-driven Autoscaling to scale pods based on Azure Service Bus queue depth rather than raw CPU spikes.
- • 3-Year Savings Plan: Locked in a 3-Year Azure Savings Plan on baseline cluster nodes for an additional 38% compute discount.