Skip to content

Resource-Level Cloud-Cost Optimization

1 · Purpose

Provide customers an automated framework to right-size compute & database instances, modernize storage, eliminate idle/orphaned assets, and tame network/data-transfer spend—continuously and at scale.

2 · Framework Overview

CloudWatch Metrics   Trusted Advisor   Cost Explorer
│                    │                │
├────────► AWS Compute Optimizer ◄────┤
│                                      │
▼                                      ▼
Optimization Hub (Athena views + QuickSight)  ◄─‒‒ Tag-driven schedules (Lambda)
│
▼
Automation Engine (SSM Automation, EventBridge, Lambda)

3 · Compute & Database Right-Sizing

Feature Implementation
Utilization metrics Import CPU, network, disk I/O, memory (CloudWatch agent) into Compute Optimizer
Cross-family recommendations Enable EC2 enhanced metrics to get suggestions for Graviton/M7i/T4g families
RDS right-sizing RDS Performance Insights + Trusted Advisor => “modify-db-instance” recommendations
Newer generations Daily script filters Optimizer output for “can be modernized” flag
Approval workflow Jira ticket auto-created; change applies via SSM Automation after CAB sign-off

Sample CLI extract

aws compute-optimizer get-ec2-instance-recommendations \
  --filter name=Finding,values=Underprovisioned \
  --query "instanceRecommendations[].[instanceArn,recommendedInstanceType]"

4 · Elasticity & Scheduling

Use Case Tactic
Office-hour dev/test stacks Instance Scheduler solution – tag Schedule=weekday9to6
Predictable nightly batch Auto Scaling scheduled actions – scale to 0 at 02:00 UTC
Ad-hoc workloads Lambda checks past seven-day CloudWatch metrics; stops idle instances after 2 h below 5 % CPU

5 · Storage Optimization

5.1 Metrics & Modernization

  • CloudWatch VolumeIdleTime, VolumeThroughputPercentage drive GP2 → GP3 list.
  • SSM Automation document AWSSupport-ModifyEBStoGP3 executes change.

5.2 Snapshot Governance

Policy Tool
Tag-based retention (Retention=30) Lambda + EventBridge (“EBS-Snapshot-Cleaner”)
Un-tagged snapshots older > 90 d Step Functions workflow requests tag or deletes

5.3 Lifecycle & Tiering

Storage Automation
S3 Intelligent-Tiering, Lifecycle rules to Glacier Deep Archive after 365 d
EFS Infrequent-Access on after 30 d idle
Incomplete multipart uploads Lifecycle rule: abort after 7 d

6 · Idle / Orphaned Resource Cleanup

Resource Detection Query Remediation
Elastic IPs DescribeAddresses where AssociationId NULL Lambda releases after 3 d
Unattached EBS vols Athena CUR query UsageType='EBS:VolumeUsage' AND ResourceId NOT IN (ec2 vols in use) SSM runbook snapshots & deletes
Idle RDS CloudWatch CPUUtilization < 3% AND ConnCount < 1 for 7 d Stop DB; notify owner tag
Idle Redshift Trusted Advisor + cluster CPU metrics Pause cluster; review after 14 d
Unused VPCs No ENIs + no routes Terraform PR deletes
Idle ALBs RequestCount < 1 7 d Lambda deletes

7 · Networking & Data-Transfer Cost Controls

  • Athena‡CUR view highlights DataTransfer-Out-Bytes spikes.
  • Enable Amazon CloudFront + Regional Edge Cache for egress heavy apps.
  • S3 → EC2 same-AZ pattern enforced by Config rule; cross-AZ flag triggers Jira.
  • VPC interface endpoints replace NAT GW where feasible; cost impact tracked monthly.

8 · Continuous Optimization Process

Frequency Task Owner Tool
Daily Pull Optimizer + Trusted Advisor findings → DynamoDB Lambda
Weekly Generate QuickSight “Top 20 Savings Ops” report FinOps Analyst
Monthly Execute approved Automation runbooks DevOps
Quarterly Re-evaluate RI/SP mix based on right-sizing results FinOps + SA

9 · Sample QuickSight “Top Savings Opportunities”

Resource Rec Type Old → New Annual Savings
i-0ab1c EC2 t3.large → t4g.medium \$ 410
mysql-prod-db RDS r5.large → r7g.large \$ 720
15 GP2 vols Convert to GP3 \$ 530

10 · Deliverables

  • Right-Sizing Report – CSV + dashboard link
  • GP2→GP3 Migration Plan – SSM Automation doc + schedule
  • Snapshot & Orphan Cleaner – Lambda code (embedded appendix)
  • Monthly Optimization Summary – PDF auto-emailed to stakeholders

Appendix – Lambda Skeleton: EBS Snapshot Cleaner

import boto3, os, datetime
ec2 = boto3.client('ec2')
retention = int(os.environ['RETENTION_DAYS'])

def lambda_handler(event, context):
    today = datetime.datetime.utcnow()
    snaps = ec2.describe_snapshots(OwnerIds=['self'])['Snapshots']
    for s in snaps:
        if 'Tags' not in s:
            age = (today - s['StartTime'].replace(tzinfo=None)).days
            if age > retention:
                ec2.delete_snapshot(SnapshotId=s['SnapshotId'])

Last updated: 30 Jun 2025