Cost Optimization Patterns

Cloud Cost Optimization

Right-Sizing

# AWS - Compute Optimizer önerileri
aws compute-optimizer get-ec2-instance-recommendations

# CPU/Memory utilization < 20% → downsize
# CPU/Memory utilization > 80% → upsize veya auto-scale

Instance Strategy

Strateji	Tasarruf	Risk	Use Case
On-Demand	0%	Düşük	Değişken workload
Reserved (1yr)	%30-40	Orta	Steady-state
Reserved (3yr)	%50-60	Yüksek	Uzun vadeli
Spot/Preemptible	%60-90	Yüksek	Batch, CI/CD
Savings Plans	%30-50	Düşük	Flexible commitment

Serverless Cost

Lambda cost = (requests × $0.20/1M) + (GB-seconds × $0.0000166)

Optimizasyon:
1. Memory right-sizing (power tuning)
2. Provisioned concurrency (cold start vs cost tradeoff)
3. ARM architecture (%20 ucuz)
4. Batch processing (SQS batch size)

LLM/API Token Cost

// Token maliyet takibi
const MODEL_COSTS = {
  'claude-opus-4-6':   { input: 15.0, output: 75.0 },   // per 1M tokens
  'claude-sonnet-4-6': { input: 3.0,  output: 15.0 },
  'claude-haiku-4-5':  { input: 0.80, output: 4.0 }
} as const

function estimateCost(model: string, inputTokens: number, outputTokens: number) {
  const costs = MODEL_COSTS[model]
  return (inputTokens * costs.input + outputTokens * costs.output) / 1_000_000
}

Build Time Optimization

Teknik	Tasarruf	Karmaşıklık
Dependency caching	%30-50	Düşük
Parallel test execution	%40-60	Düşük
Incremental builds	%50-70	Orta
Docker layer caching	%20-40	Düşük
Affected-only (monorepo)	%60-80	Orta

Database Cost

-- Pahalı query'leri bul
SELECT query, calls, mean_exec_time, total_exec_time
FROM pg_stat_statements
ORDER BY total_exec_time DESC
LIMIT 20;

-- Unused index'leri bul (gereksiz storage + write overhead)
SELECT indexrelname, idx_scan
FROM pg_stat_user_indexes
WHERE idx_scan = 0 AND indexrelname NOT LIKE '%pkey%';

Checklist

• Cloud cost alerting aktif (budget alarm)
• Unused resources temizlenmiş
• Right-sizing review (aylık)
• Reserved/Spot instance kullanımı
• CDN cache hit ratio >90%
• Database query optimization
• Build cache aktif
• Tag-based cost allocation

Anti-Patterns

• Over-provisioned "just in case" resources
• Unused EBS volumes, idle load balancers
• Data transfer between regions (pahalı)
• CloudWatch log retention unlimited
• Her environment production-grade