Cost Optimization on AWS for Java Workloads

Cost awareness is a feature. Treat it like performance: measure, set targets, iterate.

Quick Wins

• Migrate to Graviton-based instances for 20–40% better price/perf
• Right-size with autoscaling and target tracking on ALB latency
• Use EBS gp3 volumes; detach unused EBS and snapshots

JVM Tuning Essentials

• Prefer G1/Generational ZGC for low-latency APIs
• Set container-aware heap: -XX:+UseContainerSupport
• Emit structured logs; sample traces instead of full capture

Purchasing Options

• Savings Plans for steady-state compute
• Spot for stateless batch and non-prod
• S3 lifecycle rules and Intelligent-Tiering for object storage

Regular cost reviews prevent drift and keep performance predictable.

A Repeatable Monthly Cost Review (2 hours)

1. Pull Costs by Service and Tag

• Use Cost Explorer or CUR Athena to list spend by Service, LinkedAccount, Tag:env, Tag:app
• Identify top 5 services: EC2/ECS/EKS, RDS, S3, NAT, Data Transfer

2. Compute: Rightsize and Purchase Options

• Compare CPU/mem utilization vs instance sizing
• Move to Graviton where supported (Java 17/21 runs great)
• Commit Savings Plans for baseline; use Spot for non-prod/batch

3. Java Runtime Tuning

# Recommended container-aware flags
JAVA_TOOL_OPTIONS="-XX:+UseContainerSupport -XX:MaxRAMPercentage=70 -XX:+AlwaysActAsServerClassMachine"

• Prefer G1 (APIs) or Generational ZGC (low-latency)
• Emit metrics to Prometheus and watch GC pause times vs throughput

4. Storage and Data

• Switch EBS to gp3 and set throughput appropriately
• Use S3 lifecycle policies, Intelligent-Tiering for rarely accessed objects
• RDS: enable storage autoscaling, review instance class quarterly

5. Networking

• Evaluate NAT costs; move egress to VPC endpoints where possible
• Cache outbound calls; batch and compress payloads

6. CI/CD Hygiene

• Delete unused ECR images and old task definitions
• Tear down ephemeral environments automatically (TTL)

Cost Guardrails to Automate

• Budgets with alerts at 50/80/100%
• CloudWatch alarms on unusual spikes (e.g., NAT bytes, S3 GETs)
• Weekly report: cost per request, per environment, per team