65% AWS cost reduction ($3,850 → $1,330 / month) via safe legacy decommissioning, zero downtime for production systems.
The challenge
Blender Networks Inc., an advertising agency based in Bedford, Canada, had a bloated AWS bill, root cause: a legacy, proprietary ad-serving platform nobody seriously used anymore, yet still burning compute, load balancers and storage. Goal: decommission it cleanly without the production WordPress sites or email infrastructure (MX) going down for even a second.
The approach
- Log analysis, not guesswork. Deep dive into ALB access logs to separate legitimate traffic from noise, uncovered hidden dependencies between the old EKS cluster and live production and cleanly decoupled them.
- 8 "zombie ALBs" eliminated. Load balancers that served no real traffic anymore, just magnets for bot scans (PHPUnit exploits, credential stuffing). Expensive noise with no business value. Shut down.
- EKS cluster phased down. Controlled shutdown of the entire Kubernetes cluster including worker nodes; orphaned target groups and load balancers removed.
- Storage cleanup. 1.4 TB of obsolete EBS snapshots and system logs deleted. 1 TB of historical RDS backups moved to Glacier Deep Archive via an S3 lifecycle rule.
- EC2 right-sizing. Once bot traffic and legacy overhead were gone, the main server could be safely scaled down, the final push on compute cost.
Before / after: what drove the old bill and the steps that brought it down.
The result
- −65% AWS cost, from ~$3,850 to $1,330 per month.
- Zero downtime for production WordPress and email throughout the entire decommissioning.
- 1.4 TB storage freed, 1 TB migrated to cold storage.
- 8 attack vectors eliminated via zombie-ALB shutdown.
Top comments (0)