Cost Optimization
Systematically reduce infrastructure and operational costs while maintaining SLAs and performance.
Context
You are optimizing costs. The user faces high cloud bills or needs to improve unit economics. Read their current infrastructure and cost breakdown.
Domain Context
Based on AWS, Google Cloud, and Azure cost optimization practices:
- Reserved Instances: Commit to usage upfront; 30-70% discount vs on-demand
- Spot Instances: Unused capacity at 70-90% discount; risk of interruption
- Right-Sizing: Match instance type to actual usage (many over-provisioned)
- Resource Scheduling: Run infrastructure only when needed (dev, test, non-critical workloads)
- Data Transfer Costs: Moving data between regions or to internet expensive; design to minimize
Instructions
Measure Current Costs: Break down: compute (instances, pods), storage (databases, backups, archives), networking (data transfer, CDN), managed services. Identify top 3 cost drivers.
Baseline Cost per Unit: Calculate cost per request, per transaction, per user. Enables comparing optimizations objectively.
Identify Optimization Opportunities:
- Compute: Right-size instances, use reserved/spot, schedule non-critical
- Storage: Archive old data, compress backups, dedup
- Transfer: Use CDN, minimize cross-region, batch transfers
- Waste: Unused resources (old databases, unused backups, forgotten VMs)
Model Impact: For each optimization, estimate savings and implementation effort. Prioritize by ROI (savings / effort).
Implement & Monitor: Deploy changes. Measure cost reduction. Re-baseline and repeat.
Anti-Patterns
- Chasing Lowest Cost: Use cheapest options (spot instances, no redundancy). Result: outages and data loss. Guard: Optimize cost subject to SLA constraints.
- Over-Provisioning for Safety: Buy 10x capacity to "be safe." Result: wasted money. Guard: Monitor actual usage; right-size to p95 load.
- Ignoring Total Cost of Ownership: Count compute but not operations. Result: hidden costs. Guard: Include all costs (instances, storage, monitoring, support, ops time).
- One-Time Optimization: Cut costs once, never revisit. Result: costs creep back up. Guard: Quarterly cost review; treat as ongoing process.
Further Reading
- AWS Well-Architected Framework (Cost Optimization Pillar) — systematic cost reduction
- Cloud Economics by Joe Weinman — understanding cloud cost models and optimization
- Google Cloud Cost Optimization Best Practices — practical examples and case studies
1---2name: cost-optimization3description: Optimize infrastructure and operational costs without sacrificing performance or reliability. Use when managing cloud budgets or improving unit economics.4---56# Cost Optimization78Systematically reduce infrastructure and operational costs while maintaining SLAs and performance.910## Context1112You are optimizing costs. The user faces high cloud bills or needs to improve unit economics. Read their current infrastructure and cost breakdown.1314## Domain Context1516Based on AWS, Google Cloud, and Azure cost optimization practices:1718- **Reserved Instances**: Commit to usage upfront; 30-70% discount vs on-demand19- **Spot Instances**: Unused capacity at 70-90% discount; risk of interruption20- **Right-Sizing**: Match instance type to actual usage (many over-provisioned)21- **Resource Scheduling**: Run infrastructure only when needed (dev, test, non-critical workloads)22- **Data Transfer Costs**: Moving data between regions or to internet expensive; design to minimize2324## Instructions25261. **Measure Current Costs**: Break down: compute (instances, pods), storage (databases, backups, archives), networking (data transfer, CDN), managed services. Identify top 3 cost drivers.27282. **Baseline Cost per Unit**: Calculate cost per request, per transaction, per user. Enables comparing optimizations objectively.29303. **Identify Optimization Opportunities**:31 - Compute: Right-size instances, use reserved/spot, schedule non-critical32 - Storage: Archive old data, compress backups, dedup33 - Transfer: Use CDN, minimize cross-region, batch transfers34 - Waste: Unused resources (old databases, unused backups, forgotten VMs)35364. **Model Impact**: For each optimization, estimate savings and implementation effort. Prioritize by ROI (savings / effort).37385. **Implement & Monitor**: Deploy changes. Measure cost reduction. Re-baseline and repeat.3940## Anti-Patterns4142- **Chasing Lowest Cost**: Use cheapest options (spot instances, no redundancy). Result: outages and data loss. **Guard**: Optimize cost subject to SLA constraints.43- **Over-Provisioning for Safety**: Buy 10x capacity to "be safe." Result: wasted money. **Guard**: Monitor actual usage; right-size to p95 load.44- **Ignoring Total Cost of Ownership**: Count compute but not operations. Result: hidden costs. **Guard**: Include all costs (instances, storage, monitoring, support, ops time).45- **One-Time Optimization**: Cut costs once, never revisit. Result: costs creep back up. **Guard**: Quarterly cost review; treat as ongoing process.4647## Further Reading4849- _AWS Well-Architected Framework (Cost Optimization Pillar)_ — systematic cost reduction50- _Cloud Economics_ by Joe Weinman — understanding cloud cost models and optimization51- _Google Cloud Cost Optimization Best Practices_ — practical examples and case studies
sethdford