Tune AI Landing Zone for Production

When to Use

• User asks to optimize landing zone cost
• User asks about SKU sizing or right-sizing
• User asks to configure monitoring alerts
• User mentions quota, budget, cost optimization

Tuning Dimensions

Phase 1: SKU Right-Sizing

AI Services

Compute SKUs

# Check current usage vs capacity
az vm list-usage --location eastus2 \
  --query "[?contains(name.value, 'Standard_NC')].{SKU:name.localizedValue, Used:currentValue, Limit:limit}" \
  -o table

# Right-sizing analysis
# If GPU utilization < 30% → downgrade SKU
# If GPU utilization > 80% → upgrade or add nodes
# If CPU-only workload → don't use GPU SKU

Phase 2: Network Optimization

Firewall Rule Consolidation

# List all firewall rules
az network firewall network-rule collection list \
  -g rg-hub --firewall-name fw-hub -o table

# Check for redundant rules
# ❌ Multiple rules for same FQDN → consolidate
# ❌ Allow * on any port → restrict to specific ports
# ✅ Specific FQDNs + specific ports = minimal attack surface

NSG Flow Optimization

# Check NSG flow logs for unused rules
az network watcher flow-log list -l eastus2 -o table

# Rules with 0 hits in 30 days → candidates for removal
# Rules with deny hits → working correctly (keep)

Phase 3: Monitoring Thresholds

Alert Rules by Priority

Log Analytics Optimization

# Check data ingestion volume
az monitor log-analytics workspace show \
  -g rg-ai-landing-zone -n ws-ai-landing-zone \
  --query "sku.name" -o tsv

# If ingesting > 100 GB/day → switch to commitment tier
# Common waste: verbose diagnostic logs on stable resources
# Fix: Set diagnostic settings to log only errors + critical metrics

Phase 4: Quota Planning

# GPU quota per region
az vm list-usage --location eastus2 \
  --query "[?contains(name.value, 'NC') || contains(name.value, 'ND')].{SKU:name.localizedValue, Used:currentValue, Limit:limit}" \
  -o table

# OpenAI TPM quota
az cognitiveservices account deployment list \
  --name oai-landing-zone -g rg-ai-landing-zone \
  --query "[].{Model:model.name, SKU:sku.name, Capacity:sku.capacity}" \
  -o table

# Request quota increase if needed
# Portal: Subscriptions → Usage + quotas → Request increase
# Or: az support tickets create ...

Phase 5: Cost Governance

# Create budget
az consumption budget create \
  --budget-name "ai-landing-zone-monthly" \
  --amount 5000 \
  --category Cost \
  --time-grain Monthly \
  --start-date $(date -d "first day of this month" +%Y-%m-01) \
  --end-date $(date -d "first day of next year" +%Y-01-01) \
  --resource-group rg-ai-landing-zone

# Add budget alert at 80% and 100%
# Sends email to finance + engineering team

Cost Comparison by Environment

Output: Tuning Report

## Landing Zone Tuning Report
| Dimension | Before | After | Savings |
|-----------|--------|-------|---------|
| AI Search SKU | S2 ($1000/mo) | S1 ($250/mo) | -75% |
| Log retention | 365 days | 90 days | -67% data cost |
| Firewall rules | 45 rules | 22 rules | Cleaner, faster |
| GPU quota | 8 NCv3 (unused) | Released | Freed for other teams |
| Budget alerts | None | 80% + 100% | Cost visibility |
| Monitoring alerts | 0 | 6 priority alerts | Incident detection |