↓ 28%
Measured through comparative GCP billing analysis across identical workloads
Cost-Efficient Kubernetes Orchestration
Developed a cost-optimized Kubernetes orchestration system featuring a hybrid task autoscaler and custom scheduling algorithms for heterogeneous cluster environments on Google Cloud.
Team projectPublic repositoryJan 2025 – Apr 2025IIT Jodhpur, India
Impact metrics
Key outcomes delivered for stakeholders
+35%
Average CPU/memory efficiency improvement on test cluster
10+
Microservices and batch workloads balanced across nodes
Highlights
Notable milestones and system improvements
- Achieved 28% cost reduction and 35% higher utilization compared to default GKE scheduler
- Implemented fault-tolerant scheduling logic resilient to preemptible VM churn
- Enabled per-pod telemetry collection with Prometheus for real-time scaling insight
Responsibilities
Where I created the most impact
- Built a custom Kubernetes scheduler and autoscaler (HTAS) leveraging CRDs and event-driven scaling for GKE clusters
- Designed the Resource Profiler for dynamic workload characterization and Task Packer using BFD/TBFD bin-packing algorithms
- Benchmarked performance through synthetic and real workloads, tuning placement heuristics for compute heterogeneity
Project narrative
Snapshot of the project background, execution, and results
Project context
High GKE costs from inefficient default scheduling motivated the need for custom orchestration tuned for mixed workloads and preemptible resources.
Approach
Devised a modular Kubernetes extension with CRDs and autoscaling logic driven by real-time resource profiling, integrating seamlessly with GCP’s managed cluster APIs.
Impact
Delivered a cost-conscious orchestration platform demonstrating tangible infrastructure savings and improved utilization, informing future IITJ cloud infrastructure research.
Stack
Tools and frameworks that powered the build
KubernetesGCPPythonAutoscalingCloud Optimization