S

SaaS Provider

SaaS

Case Study

58% Kubernetes Cost Reduction for a SaaS Provider

Optimized a SaaS provider's Kubernetes infrastructure, reducing cluster costs by 58% while improving reliability and performance.

KubernetesKarpenterVPAKubecostPrometheusFinOps
58%
Kubernetes cost reduction
60+
Workloads right-sized
Improved
Reliability and pod stability
90 days
Optimisation timeline

The Challenge

The client ran multiple Kubernetes clusters with static node pools sized for peak load. Off-peak hours — roughly 14 hours per day — saw utilisation below 15%, but costs remained constant. The engineering team lacked visibility into per-service resource consumption and had no autoscaling configured.

Our Solution

KubeAce implemented Karpenter for node-level autoscaling, tuned resource requests and limits across 60+ workloads based on actual utilisation data, introduced Vertical Pod Autoscaler recommendations, consolidated underutilised clusters, and established resource governance policies to prevent future over-provisioning.

Overview

A B2B SaaS company running on AWS EKS had seen Kubernetes costs grow steadily as the team added services without corresponding resource governance. Cluster costs were the single largest line item in their infrastructure budget.

Root Cause Analysis

KubeAce ran a 30-day resource profiling exercise before making any changes:

  • Resource requests vs. actual usage: average pod requested 4× its actual CPU consumption
  • Memory limits: set arbitrarily — 70% of pods had never reached 40% of their limit
  • Node utilisation: average 18% CPU, 24% memory across all nodes
  • No autoscaling: fixed node counts regardless of time-of-day load patterns

Optimisation Actions

Node-Level

  • Replaced managed node groups with Karpenter NodePools
  • Configured consolidation policies: nodes scale to zero during off-peak windows
  • Mixed Spot/On-Demand: 70% Spot for stateless workloads, On-Demand only for stateful services

Workload-Level

  • Right-sized resource requests for 63 deployments based on P95 usage metrics
  • Enabled VPA in recommendation mode for ongoing request tuning
  • Set LimitRange policies per namespace to prevent unbounded resource requests

Governance

  • Kubecost deployed with per-team showback dashboards
  • ResourceQuota applied to all namespaces
  • Resource efficiency gates added to CI/CD pipeline

Results

The 58% cost reduction came primarily from Karpenter consolidation (40% of savings) and right-sizing (18% of savings). Pod stability improved as containers were no longer competing for over-committed node resources.

More Case Studies

View All →
Free 30-Minute Strategy Session — No Commitment

Ready to Transform
Your Infrastructure?

Whether you're migrating to Kubernetes, scaling a LiveKit deployment, or building a DevOps platform from scratch — our engineers have done it before.

Schedule a Free Call View Case Studies
Response within 4 hours
NDA available on request
Serve clients globally from Bangalore