Documentation

Azure SQL Optimization

Jetscale provides AI-powered cost optimization for Azure SQL Database, including single databases, elastic pools, and Azure SQL Managed Instance. Our specialized agents analyze...

Imported documentation under review

This page preserves content from the previous documentation site for migration review. Product details and public claims have not yet been revalidated.

Jetscale provides AI-powered cost optimization for Azure SQL Database, including single databases, elastic pools, and Azure SQL Managed Instance. Our specialized agents analyze your database workloads to identify right-sizing opportunities and configuration improvements.

Overview

Jetscale optimizes Azure SQL resources by analyzing:

Database utilization: DTU/vCore usage, CPU, memory, and I/O patterns
Cost analysis: Current spending vs. optimal configuration
Performance metrics: Query performance, connection patterns, storage usage
Service tier alignment: Basic, Standard, Premium, Business Critical, Hyperscale

Supported Azure SQL Types

Azure SQL Database (Single Database)

Single databases are our primary optimization target.

Database Architecture:

SqlDatabase (Optimization Target, Billable)
├── Compute (DTU or vCore)
├── Storage (Data + Log)
└── Backup Storage

What We Optimize:

Service tier: Right-size between Basic, Standard, Premium, Business Critical, Hyperscale
Compute tier: Provisioned vs. Serverless
DTU/vCore sizing: Match actual workload requirements
Storage configuration: Data max size and performance tier

Azure SQL Elastic Pools

Elastic pools share resources across multiple databases.

Pool Architecture:

SqlElasticPool (Optimization Target, Billable)
├── Database 1 - Shares pool resources
├── Database 2 - Shares pool resources
└── Database N - Shares pool resources

What We Optimize:

Pool sizing: Total DTUs/vCores for all databases
Service tier: Standard, Premium, Business Critical
Database density: Optimal number of databases per pool

Azure SQL Managed Instance

Fully managed SQL Server instance.

Instance Architecture:

SqlManagedInstance (Optimization Target, Billable)
├── Compute (vCores)
├── Storage
└── Multiple Databases

What We Optimize:

Instance sizing: vCore count (4, 8, 16, 24, 32, 40, 64, 80)
Service tier: General Purpose vs. Business Critical
Storage configuration: Data and log storage sizing

Service Tiers

DTU-Based Model (Single Database & Elastic Pools)

Basic Tier

Use for: Small databases, development, testing, light workloads
DTU range: 5 DTUs
Max database size: 2 GB
Typical cost: ~$5/month
Best when: Learning, prototyping, minimal production use

Standard Tier

Use for: Most production workloads, web applications, line-of-business apps
DTU range: 10, 20, 50, 100, 200, 400, 800, 1600, 3000 DTUs
Max database size: Up to 1 TB
Typical cost: $15-$1,500/month
Best when: Predictable performance requirements, balanced workloads

Premium Tier

Use for: Mission-critical applications, high transaction rate, low latency
DTU range: 125, 250, 500, 1000, 1750, 4000 DTUs
Max database size: Up to 4 TB
Typical cost: $465-$14,500/month
Best when: Highest performance, availability, and recovery requirements

vCore-Based Model (All Types)

General Purpose

Use for: Most production workloads, balanced compute and I/O
vCores: 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 24, 32, 40, 80, 128
Max memory: 625 GB (128 vCore)
Storage: Remote storage (SSD), up to 4 TB (single) or 16 TB (managed instance)
Typical cost: $300-$8,000/month
Best when: Standard business workloads

Business Critical

Use for: Mission-critical applications, low latency, high availability
vCores: Same range as General Purpose
Max memory: 625 GB (128 vCore)
Storage: Local SSD, up to 4 TB (single) or 16 TB (managed instance)
Built-in HA: Always-on availability group with readable secondary replicas
Typical cost: 2.5x General Purpose pricing
Best when: Highest performance and availability required

Hyperscale (Single Database only)

Use for: Very large databases (100 TB+), rapid scale-up/down
vCores: 2-80 vCores
Storage: Highly scalable, up to 100 TB
Architecture: Multi-tier with page servers and compute nodes
Typical cost: Similar to General Purpose, storage charged separately
Best when: Database > 4 TB, need rapid scaling

Compute Tiers

Provisioned

Billing: Per second for compute resources
Best for: Predictable workloads, continuous usage
Minimum: No minimum compute

Serverless (General Purpose only)

Billing: Per second when active, auto-pause when idle
Best for: Intermittent, unpredictable usage patterns
Auto-pause: Configurable delay (1 hour - 7 days)
Savings: Up to 90% during idle periods
Note: Cold start delay when resuming from paused state

How Jetscale Optimizes Azure SQL

1. Data Collection

Jetscale analyzes multiple data sources:

Azure Monitor Metrics (14-day rolling window):

cpu_percent - CPU utilization
dtu_consumption_percent - DTU usage (DTU model)
storage_percent - Storage utilization
connection_successful / connection_failed - Connection patterns
sessions_percent - Active sessions
workers_percent - Worker thread usage
deadlock - Deadlock occurrences
log_write_percent - Transaction log write usage

Azure Resource Manager Data:

Service tier and compute tier
DTU/vCore configuration
Current storage size and max size
Geo-replication status
Backup retention settings

Cost Management Data:

Current monthly spend per database
Compute, storage, and backup costs breakdown
Historical cost trends

Azure Advisor (if available):

Azure-generated right-sizing recommendations
Performance optimization suggestions

2. Analysis

Our AI agents perform deep analysis:

Utilization Patterns:

Peak vs. average DTU/vCore usage
Time-of-day patterns (identify idle periods for serverless)
Day-of-week patterns (weekend vs. weekday load)
Growth trends over time

Performance Assessment:

Query performance metrics
Connection patterns and failures
Deadlock frequency
Transaction log pressure

Cost Modeling:

Current cost breakdown (compute, storage, backup)
Projected cost for alternative configurations
Provisioned vs. Serverless cost comparison
Reserved capacity opportunities

Tier Selection:

Basic vs. Standard vs. Premium alignment
General Purpose vs. Business Critical requirements
Hyperscale suitability for large databases
DTU vs. vCore model efficiency

3. Recommendations

Jetscale generates specific, actionable recommendations:

Service Tier Downgrade

Example Recommendation:

Resource: app-database-prod
Current: Premium P2 (250 DTUs, $930/month)
Recommended: Standard S3 (100 DTUs, $300/month)

Cost Impact:
- Current: $930/month
- Projected: $300/month
- Savings: $630/month (68%), $7,560/year

Performance Analysis:
- Average DTU usage: 35%
- Peak DTU usage: 58%
- Connection patterns: Normal
- Query performance: Well within Standard tier capabilities

Risk: Low - Peak usage well below new tier limits
Note: Premium tier features (lower latency, more IOPS) not fully utilized

DTU Right-Sizing

Example Recommendation:

Resource: customer-db-01
Current: Standard S7 (800 DTUs, $600/month)
Recommended: Standard S3 (100 DTUs, $300/month)

Cost Impact:
- Current: $600/month
- Projected: $300/month
- Savings: $300/month (50%), $3,600/year

Performance Analysis:
- Average DTU usage: 18%
- Peak DTU usage: 42%
- 95th percentile: 35%
- Recommended tier provides 2.4x headroom above peak

Risk: Very Low - Substantial headroom maintained

Serverless Migration

Example Recommendation:

Resource: reporting-database
Current: General Purpose (4 vCores, Provisioned, $730/month)
Recommended: General Purpose (4 vCores, Serverless, ~$290/month)

Cost Impact:
- Current: $730/month (24/7 billing)
- Projected: $290/month (avg 40% active time)
- Savings: $440/month (60%), $5,280/year

Usage Analysis:
- Active hours: ~10 hours/day (business hours only)
- Idle time: Nights and weekends (60% of time)
- Query patterns: Batch processing and reporting
- Cold start tolerance: Acceptable (reporting use case)

Configuration:
- Auto-pause delay: 1 hour
- Min vCores: 1 (when active)
- Max vCores: 4 (during peak)

Risk: Very Low - Intermittent usage pattern ideal for serverless
Note: ~2-minute resume time from paused state

vCore Optimization

Example Recommendation:

Resource: api-database-prod
Current: General Purpose (16 vCores, $1,460/month)
Recommended: General Purpose (8 vCores, $730/month)

Cost Impact:
- Current: $1,460/month
- Projected: $730/month
- Savings: $730/month (50%), $8,760/year

Performance Analysis:
- Average CPU: 22%
- Peak CPU: 38%
- Memory usage: 45%
- I/O patterns: Well within General Purpose limits

Risk: Low - 2.6x headroom above peak CPU usage

Business Critical → General Purpose

Example Recommendation:

Resource: legacy-app-database
Current: Business Critical (8 vCores, $3,650/month)
Recommended: General Purpose (8 vCores, $1,460/month)

Cost Impact:
- Current: $3,650/month
- Projected: $1,460/month
- Savings: $2,190/month (60%), $26,280/year

Workload Analysis:
- Business Critical features not utilized:
  × No readable secondary replicas in use
  × Low latency requirements not critical
  × Local SSD performance not required
- Performance requirements met by General Purpose

Risk: Medium - Verify application tolerates remote storage latency
Recommendation: Test in staging environment first

Elastic Pool Consolidation

Example Recommendation:

Resource: 12 Standard S3 databases
Current: 12 × S3 (100 DTUs each, $3,600/month total)
Recommended: 1 × Elastic Pool (600 eDTUs, $900/month)

Cost Impact:
- Current: $3,600/month (12 individual databases)
- Projected: $900/month (elastic pool with 12 databases)
- Savings: $2,700/month (75%), $32,400/year

Usage Analysis:
- Individual database peak usage: 40-60 DTUs
- Simultaneous peak usage: Rare (time-shifted workloads)
- Total concurrent DTU usage: ~300 DTUs peak
- Pool provides 600 eDTUs with resource sharing

Risk: Low - Sufficient pool capacity for peak concurrent usage
Note: Databases share pool resources, reducing over-provisioning

4. Terraform Generation

For each recommendation, Jetscale generates production-ready Terraform code:

Example: Service Tier Optimization

# Azure SQL Database Optimization
# Generated by Jetscale on 2024-01-15
# Recommendation ID: rec_azuresql_001

resource "azurerm_mssql_database" "app_database_prod" {
  name      = "app-database-prod"
  server_id = azurerm_mssql_server.example.id

  # Previous: Premium P2 (250 DTUs, $930/month)
  # Optimized: Standard S3 (100 DTUs, $300/month)
  # Cost Reduction: 68% ($630/month, $7,560/year)
  #
  # Justification:
  # - Average DTU usage: 35%
  # - Peak DTU usage: 58%
  # - Premium tier features (lower latency, higher IOPS) not fully utilized
  # - Standard S3 provides adequate performance for workload
  sku_name = "S3"

  max_size_gb = 250

  # Existing configuration preserved
  collation               = "SQL_Latin1_General_CP1_CI_AS"
  zone_redundant          = false
  read_scale              = false
  storage_account_type    = "Geo"

  threat_detection_policy {
    state                      = "Enabled"
    email_account_admins       = "Enabled"
    retention_days             = 30
  }

  tags = merge(
    var.tags,
    {
      "jetscale:optimized"      = "true"
      "jetscale:recommendation" = "rec_azuresql_001"
      "jetscale:previous_sku"   = "P2"
    }
  )
}

Example: Serverless Migration

# Azure SQL Serverless Migration
# Generated by Jetscale on 2024-01-15

resource "azurerm_mssql_database" "reporting_database" {
  name      = "reporting-database"
  server_id = azurerm_mssql_server.example.id

  # Previous: Provisioned (4 vCores, 24/7, $730/month)
  # Optimized: Serverless (1-4 vCores, ~40% active, $290/month)
  # Cost Reduction: 60% ($440/month, $5,280/year)
  #
  # Usage Pattern Analysis:
  # - Active: Business hours only (~10 hrs/day)
  # - Idle: Nights and weekends (60% of time)
  # - Use case: Batch reporting (cold start acceptable)
  #
  # Serverless Configuration:
  # - Auto-pause after 1 hour of inactivity
  # - Min 1 vCore when active, max 4 vCores
  # - ~2-minute resume time from paused state
  sku_name = "GP_S_Gen5_4"

  min_capacity = 1.0
  max_capacity = 4.0

  # Auto-pause after 1 hour (60 minutes) of inactivity
  auto_pause_delay_in_minutes = 60

  max_size_gb = 100

  collation               = "SQL_Latin1_General_CP1_CI_AS"
  zone_redundant          = false
  read_scale              = false
  storage_account_type    = "Geo"

  tags = merge(
    var.tags,
    {
      "jetscale:optimized"     = "true"
      "jetscale:compute_model" = "serverless"
      "jetscale:previous_sku"  = "GP_Gen5_4"
    }
  )
}

Example: Elastic Pool

# Azure SQL Elastic Pool Consolidation
# Generated by Jetscale on 2024-01-15

resource "azurerm_mssql_elasticpool" "app_pool" {
  name                = "app-elastic-pool"
  resource_group_name = var.resource_group_name
  location            = var.location
  server_name         = azurerm_mssql_server.example.name

  # Previous: 12 individual S3 databases (1200 DTUs total, $3,600/month)
  # Optimized: Elastic pool (600 eDTUs shared, $900/month)
  # Cost Reduction: 75% ($2,700/month, $32,400/year)
  #
  # Analysis:
  # - Individual database peaks rarely coincide
  # - Peak concurrent usage: ~300 DTUs
  # - Pool provides 600 eDTUs with 2x headroom
  # - Resource sharing eliminates over-provisioning
  sku {
    name     = "StandardPool"
    tier     = "Standard"
    capacity = 600
  }

  per_database_settings {
    min_capacity = 10
    max_capacity = 100
  }

  max_size_gb = 750

  tags = merge(
    var.tags,
    {
      "jetscale:optimized"         = "true"
      "jetscale:pool_type"         = "consolidation"
      "jetscale:database_count"    = "12"
      "jetscale:previous_total_cost" = "$3600"
    }
  )
}

# Move databases to elastic pool
resource "azurerm_mssql_database" "pooled_databases" {
  count = 12

  name           = "app-database-${count.index + 1}"
  server_id      = azurerm_mssql_server.example.id
  elastic_pool_id = azurerm_mssql_elasticpool.app_pool.id

  # Databases now share pool resources
  # No individual sku_name needed

  tags = var.tags
}

Best Practices

Monitoring After Changes

After applying Jetscale recommendations:

First 24 Hours: Monitor key metrics closely
- CPU and DTU/vCore utilization
- Connection success rate
- Query performance (execution time)
- Transaction log usage
Week 1: Validate performance
- Compare query execution times
- Check for resource pressure indicators
- Monitor connection failures
- Review application error rates
Week 2-4: Confirm cost savings
- Verify Azure billing reflects expected savings
- Check for unexpected charges
- Consider reserved capacity for stable workloads

Testing Strategy

Pre-Production Testing:

Apply changes to dev/staging environment first
Run load tests simulating peak usage
Monitor for 48-72 hours under realistic load
Validate backup/restore procedures

Production Rollout:

Schedule changes during maintenance windows
Use Azure SQL geo-replication for blue/green deployments
Have rollback plan ready
Monitor actively during and after change

Reserved Capacity

When to Purchase:

Stable, long-running databases (> 1 year)
After right-sizing (don't reserve oversized resources)
Up to 80% savings vs. pay-as-you-go

Jetscale Recommendations:

1-year or 3-year commitment
Applied automatically to matching resources
Size-flexible within same service tier

Common Optimization Patterns

Pattern 1: Over-Provisioned Premium Database

Symptoms:

Premium tier with low DTU usage (<40%)
High availability features not utilized
Lower latency not critical for application

Jetscale Recommendation:

Downgrade to Standard tier
Typical savings: 60-70%
Risk: Low (verify latency requirements)

Pattern 2: Idle Development Databases

Symptoms:

Databases used only during business hours
Zero activity nights and weekends
Development or reporting workloads

Jetscale Recommendation:

Migrate to Serverless compute tier
Typical savings: 50-80%
Risk: Very Low (non-continuous usage)

Pattern 3: Multiple Small Databases

Symptoms:

Many databases with similar usage patterns
Individual database peaks don't coincide
Over-provisioning due to peak capacity per database

Jetscale Recommendation:

Consolidate into Elastic Pool
Typical savings: 60-75%
Risk: Low (resource sharing reduces waste)

Pattern 4: Business Critical Over-Provisioning

Symptoms:

Business Critical tier without HA requirements
Readable secondary replicas not used
Local SSD performance not required

Jetscale Recommendation:

Downgrade to General Purpose
Typical savings: 60% (2.5x price difference)
Risk: Medium (verify latency tolerance)

Troubleshooting

Recommendation Concerns

Q: Will downgrading impact performance?

A: Jetscale maintains 2-3x headroom above peak usage. We analyze:

P95 and P99 DTU/vCore usage over 14 days
Query performance metrics
Connection patterns and failures
Historical spike patterns

Recommendations only proceed if performance risk is Low or Very Low.

Q: What about serverless cold start delays?

A: Serverless databases resume in ~2 minutes from paused state. Best for:

Batch processing and reporting
Development and testing
Applications tolerant of occasional cold starts
Avoid for: Real-time APIs, always-on applications

Q: Can I test before committing?

A: Yes! Apply changes to dev/staging first:

Jetscale generates Terraform for all environments
Test in non-production for 48-72 hours
Monitor performance metrics
Rollback if needed (simple tier change)
Apply to production once validated

Performance Issues After Optimization

Symptom: Increased query latency

Possible causes:

Insufficient DTU/vCore capacity
Storage I/O throttling
Connection pool exhaustion

Resolution:

Check cpu_percent and dtu_consumption_percent metrics
Review slow query logs
Increase tier/size by one step if needed
Consider Query Performance Insights for optimization

Symptom: Connection failures

Possible causes:

Serverless database in paused state
Connection pool settings need adjustment
Resource limits exceeded

Resolution:

If serverless: Reduce auto-pause delay or switch to provisioned
Check connection_failed metric
Review connection pool configuration
Monitor sessions_percent for capacity

Security Considerations

Encryption

Jetscale recommendations preserve:

Transparent Data Encryption (TDE) settings
Always Encrypted configuration
SSL/TLS connection requirements
Encryption at rest and in transit

Compliance

Configuration changes maintain compliance:

Geo-replication settings preserved
Backup retention policies unchanged
Threat detection and auditing settings maintained
Virtual network rules and firewall rules preserved

Limitations

Not Currently Supported:

Reserved capacity purchase recommendations
Cross-region replication optimization
Backup retention policy optimization
Index and query tuning recommendations

Roadmap: These features are planned for future releases. Currently, Jetscale focuses on service tier and compute optimization.

API Integration

Jetscale provides API access for programmatic optimization:

# List Azure SQL recommendations
GET /api/v1/recommendations?resource_type=azure-sql

# Get specific recommendation details
GET /api/v1/recommendations/{recommendation_id}

# Approve recommendation (generates Terraform)
POST /api/v1/recommendations/{recommendation_id}/approve

# Retrieve generated Terraform
GET /api/v1/recommendations/{recommendation_id}/terraform

See our API Documentation for complete reference.

Support

Need help with Azure SQL optimization?

Email: support@jetscale.ai
Documentation: FAQ
GitHub Issues: Report a problem

Related Documentation:

S3 Optimization

Jetscale provides AI-powered cost optimization for Amazon S3 buckets. Our specialized agents analyze your storage workloads to identify tiering opportunities, lifecycle policy...

Azure VM Optimization

Jetscale provides AI-powered cost optimization for Azure Virtual Machines. Our specialized agents analyze your compute workloads to identify right-sizing opportunities, VM...