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 family migrations, and cost-effective alternatives.

Overview

Jetscale optimizes Azure VM resources by analyzing:

• Instance utilization: CPU, memory, disk, and network usage patterns
• Cost analysis: Current spending vs. optimal configuration
• Workload characteristics: Resource requirements and performance patterns
• Pricing options: Pay-as-you-go, Reserved Instances, Spot VMs

Supported VM Types

Standalone Virtual Machines

Jetscale optimizes standalone Azure VMs that are not part of Virtual Machine Scale Sets (VMSS).

VM Architecture:

VirtualMachine (Optimization Target, Billable)
├── OSDisk - Billable
├── DataDisk - Billable (optimized separately)
└── DataDisk - Billable

What We Optimize:

• VM size right-sizing: Change to smaller/larger sizes based on actual usage
• VM series migration: Switch between series (B, D, E, F, etc.) based on workload characteristics
• Generation upgrades: Migrate to latest generation (v5, v6) for better price/performance
• Delete unused: Remove VMs with zero or near-zero activity (100% savings)

Important: Jetscale currently optimizes standalone Azure VMs only. VMs managed by Virtual Machine Scale Sets, AKS, or other orchestration systems are not yet supported.

VM Series and Families

Jetscale considers all Azure VM series when optimizing:

General Purpose (B, D)

B-series - Burstable VMs

• Use for: Dev/test, low-traffic web servers, small databases, microservices
• Best when: CPU usage is low most of the time with occasional bursts
• Caution: Based on CPU credits, monitor credit balance to avoid throttling
• Sizes: B1s, B1ms, B2s, B2ms, B4ms, B8ms, B12ms, B16ms, B20ms

D-series - Balanced compute/memory/network

• Use for: Web servers, application servers, medium databases, mixed workloads
• Latest generations: Dv5/Dsv5 > Dv4/Dsv4 > Dv3/Dsv3
• Available in: Standard HDD, Standard SSD, Premium SSD
• Sizes: D2-D96 (2-96 vCPUs)

Compute Optimized (F)

F-series - High CPU-to-memory ratio

• Use for: Batch processing, application servers, web servers, analytics, gaming
• Best when: CPU utilization is consistently high, memory needs are moderate
• Latest generations: Fv2 (Intel), Fasv5/Famsv5 (AMD)
• Sizes: F2-F72 (2-72 vCPUs)

Memory Optimized (E, M)

E-series - High memory-to-CPU ratio

• Use for: Relational databases, in-memory analytics, SAP applications, caching
• Best when: Memory utilization is high, CPU is moderate
• Latest generations: Ev5/Esv5 > Ev4/Esv4 > Ev3/Esv3
• Sizes: E2-E96 (2-96 vCPUs, up to 672 GB RAM)

M-series - Extreme memory (up to 4 TB RAM)

• Use for: Large SQL Server, SAP HANA, in-memory databases
• Cost premium: Significantly higher than E-series
• Mission-critical workloads requiring massive memory
• Sizes: M8ms - M416ms (8-416 vCPUs)

Storage Optimized (L)

L-series - High disk throughput and I/O

• Use for: Big data, SQL databases, NoSQL databases, data warehousing
• NVMe-based local storage
• Best for: I/O-intensive workloads requiring low latency
• Sizes: L4s-L80s (4-80 vCPUs)

Accelerated Computing (N)

N-series - GPU-enabled VMs

• Use for: AI/ML training and inference, graphics rendering, video processing
• Options: NC (NVIDIA Tesla), NV (NVIDIA Tesla M60), ND (NVIDIA Tesla P40/V100)
• Not typically recommended for general cost optimization (specialized workloads)

VM Sizing Patterns

Sizes follow pattern: {Series}{Version}{Size}

Examples:

• B2s: B-series, 2 vCPUs, Standard disk support
• D4s_v5: D-series v5, 4 vCPUs, Premium SSD support
• E8as_v5: E-series v5 AMD, 8 vCPUs, Premium SSD support

Size progression:

• Standard: 1, 2, 4, 8, 16, 32, 64, 96+ vCPUs
• Burstable: 1, 2, 4, 8, 12, 16, 20 vCPUs

Each step typically doubles vCPU and memory.

How Jetscale Optimizes Azure VMs

1. Data Collection

Jetscale analyzes multiple data sources:

Azure Monitor Metrics (14-day rolling window):

• Percentage CPU - VM CPU usage
• Available Memory Bytes - Available memory
• Network In Total / Network Out Total - Network throughput
• Disk Read Bytes / Disk Write Bytes - Disk I/O
• Disk Read Operations/Sec / Disk Write Operations/Sec - IOPS

Azure Resource Manager Data:

• VM size and state
• OS type (Windows/Linux)
• Creation time
• Location and availability zone
• Tags and resource groups

Cost Management Data:

• Current monthly spend per VM
• Historical cost trends
• Reserved Instance coverage

Azure Advisor (if available):

• Azure-generated right-sizing recommendations
• Performance risk assessments

2. Analysis

Our AI agents perform deep analysis:

Utilization Patterns:

• Peak vs. average CPU utilization
• Time-of-day patterns (identify idle periods)
• Day-of-week patterns (weekend vs. weekday load)
• Memory usage trends

Workload Classification:

• Variable CPU: Consider B-series (burstable)
• Steady high CPU: Consider F-series (compute optimized)
• Balanced: Consider D-series (general purpose)
• Memory-intensive: Consider E-series (memory optimized)

Cost Modeling:

• Current monthly cost
• Projected cost for alternative VM sizes
• Savings percentage and annual impact
• Reserved Instance opportunities

Risk Evaluation:

• Performance degradation probability
• Headroom calculation (buffer above peak usage)
• Memory safety checks
• Network and storage capacity validation

3. Recommendations

Jetscale generates specific, actionable recommendations:

VM Right-Sizing

Example Recommendation:

Resource: web-server-prod-01
Current: Standard_D16s_v5 (16 vCPUs, 64 GB RAM)
Recommended: Standard_D8s_v5 (8 vCPUs, 32 GB RAM)

Cost Impact:
- Current: $550/month
- Projected: $275/month
- Savings: $275/month (50%), $3,300/year

Performance Analysis:
- Average CPU: 18%
- Peak CPU: 32%
- Average Memory: 35%
- Recommended provides 2x headroom above peak

Risk: Low - Ample headroom maintained

Series Migration

Example Recommendation:

Resource: batch-processor-02
Current: Standard_D8s_v5 (8 vCPUs, 32 GB RAM)
Recommended: Standard_F8s_v2 (8 vCPUs, 16 GB RAM)

Cost Impact:
- Current: $275/month
- Projected: $233/month
- Savings: $42/month (15%), $504/year

Workload Analysis:
- Average CPU: 75%
- Peak CPU: 92%
- Memory usage: Low (28%)
- Workload type: Compute-intensive, memory underutilized

Recommendation: Migrate to F-series (compute optimized)
Risk: Very Low - CPU-bound workload, memory headroom adequate

Generation Upgrade

Example Recommendation:

Resource: api-server-01
Current: Standard_D4s_v3 (4 vCPUs, 16 GB RAM)
Recommended: Standard_D4s_v5 (4 vCPUs, 16 GB RAM)

Cost Impact:
- Current: $175/month
- Projected: $165/month
- Savings: $10/month (6%), $120/year

Performance Analysis:
- v5 offers better CPU performance (Intel Xeon Platinum 8370C)
- Same specifications, lower cost
- Better network performance (up to 12.5 Gbps)
- Newer generation efficiency

Risk: Very Low - Same series, newer generation

Delete Unused VM

Example Recommendation:

Resource: legacy-test-vm
Current: Standard_B2s (2 vCPUs, 4 GB RAM)
Recommended: Delete/Deallocate

Cost Impact:
- Current: $30/month
- Projected: $0/month
- Savings: $30/month (100%), $360/year

Usage Analysis:
- Average CPU: 0.3%
- Network activity: Near zero
- Created: 14 months ago
- Tags indicate: test environment

Risk: Very Low - Appears unused
Recommendation: Verify with application team before deletion

4. Terraform Generation

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

Example: VM Right-Sizing

# Azure VM Optimization
# Generated by Jetscale on 2024-01-15
# Recommendation ID: rec_azurevm_001

resource "azurerm_linux_virtual_machine" "web_server_prod_01" {
  name                = "web-server-prod-01"
  resource_group_name = var.resource_group_name
  location            = var.location

  # Previous: Standard_D16s_v5 ($550/month)
  # Optimized: Standard_D8s_v5 ($275/month)
  # Cost Reduction: 50% ($275/month, $3,300/year)
  #
  # Justification:
  # - Average CPU utilization: 18%
  # - Peak CPU utilization: 32%
  # - Average memory utilization: 35%
  # - New size provides 2x headroom above peak usage
  size = "Standard_D8s_v5"

  admin_username = var.admin_username

  admin_ssh_key {
    username   = var.admin_username
    public_key = var.ssh_public_key
  }

  network_interface_ids = [
    azurerm_network_interface.web_server_nic.id,
  ]

  os_disk {
    caching              = "ReadWrite"
    storage_account_type = "Premium_LRS"
    disk_size_gb         = 128
  }

  source_image_reference {
    publisher = "Canonical"
    offer     = "0001-com-ubuntu-server-focal"
    sku       = "20_04-lts-gen2"
    version   = "latest"
  }

  tags = merge(
    var.tags,
    {
      "jetscale:optimized"      = "true"
      "jetscale:recommendation" = "rec_azurevm_001"
      "jetscale:previous_size"  = "Standard_D16s_v5"
    }
  )

  lifecycle {
    create_before_destroy = true
  }
}

Example: Series Migration

# Azure VM Series Migration
# Generated by Jetscale on 2024-01-15

resource "azurerm_linux_virtual_machine" "batch_processor_02" {
  name                = "batch-processor-02"
  resource_group_name = var.resource_group_name
  location            = var.location

  # Previous: Standard_D8s_v5 (General Purpose, $275/month)
  # Optimized: Standard_F8s_v2 (Compute Optimized, $233/month)
  # Cost Reduction: 15% ($42/month, $504/year)
  #
  # Workload Analysis:
  # - Average CPU: 75%, Peak CPU: 92%
  # - Memory usage: 28% (underutilized)
  # - Workload type: Compute-intensive
  # - F-series provides higher CPU-to-memory ratio
  size = "Standard_F8s_v2"

  admin_username = var.admin_username

  admin_ssh_key {
    username   = var.admin_username
    public_key = var.ssh_public_key
  }

  network_interface_ids = [
    azurerm_network_interface.batch_processor_nic.id,
  ]

  os_disk {
    caching              = "ReadWrite"
    storage_account_type = "Premium_LRS"
    disk_size_gb         = 256
  }

  source_image_reference {
    publisher = "Canonical"
    offer     = "0001-com-ubuntu-server-focal"
    sku       = "20_04-lts-gen2"
    version   = "latest"
  }

  tags = merge(
    var.tags,
    {
      "jetscale:optimized"     = "true"
      "jetscale:series"        = "compute-optimized"
      "jetscale:previous_size" = "Standard_D8s_v5"
    }
  )

  lifecycle {
    create_before_destroy = true
  }
}

Best Practices

Testing Strategy

Pre-Production Testing:

1. Apply changes to dev/staging environment first
2. Run load tests simulating peak traffic
3. Monitor for 48-72 hours under realistic load
4. Validate application performance

Production Rollout:

1. Use create_before_destroy lifecycle in Terraform
2. Schedule changes during maintenance windows
3. Have rollback plan ready (previous Terraform state)
4. Monitor actively during and after change

Reserved Instances & Savings Plans

While Jetscale doesn't currently generate RI recommendations, consider them after right-sizing:

Best practice:

1. Right-size VMs with Jetscale first
2. Run optimized configuration for 30-60 days
3. Purchase Reserved Instances for stable workloads (1-year or 3-year)
4. Never reserve over-sized VMs

Azure Reserved VM Instances:

• Up to 72% savings vs. pay-as-you-go
• 1-year or 3-year commitment
• Size flexibility within same VM series
• Payment options: All upfront, Partial upfront, Monthly

Spot VMs

For non-critical workloads, consider Azure Spot VMs:

• Up to 90% savings vs. pay-as-you-go
• Best for: Batch processing, dev/test, stateless applications
• Risk: Can be evicted with 30-second notice when Azure needs capacity
• Jetscale identifies candidates for Spot migration (roadmap feature)

Common Optimization Patterns

Pattern 1: Over-Provisioned VMs

Symptoms:

• CPU utilization < 20% average
• Memory utilization < 30%
• VM running 24/7 with consistent low usage

Jetscale Recommendation:

• Downsize by 1-2 VM sizes (e.g., D16 → D8)
• Typical savings: 40-50%
• Risk: Low (2-3x headroom maintained)

Pattern 2: Wrong VM Series

Symptoms:

• Using D-series with consistently high CPU (>70%)
• Using E-series with low memory usage (<40%)
• Using F-series with high memory needs

Jetscale Recommendation:

• Migrate to appropriate series (D → F for CPU, E → D for balanced)
• Typical savings: 10-30%
• Risk: Low to Medium (verify workload characteristics)

Pattern 3: Legacy Generations

Symptoms:

• Using v3 or v4 VMs
• Newer generations available (v5, v6)
• Same cost or lower for better performance

Jetscale Recommendation:

• Upgrade to latest generation (Dv3 → Dv5, Ev4 → Ev5)
• Typical savings: 5-15% with better performance
• Risk: Very Low (same series, newer hardware)

Pattern 4: Idle Development VMs

Symptoms:

• Dev/test VMs running 24/7
• Used only during business hours
• Low CPU usage overnight and weekends

Jetscale Recommendation:

• Implement auto-shutdown schedules (Azure feature)
• Downsize to B-series for burstable workloads
• Typical savings: 50-70% combined approach
• Risk: Very Low (non-production environment)

Troubleshooting

Recommendation Concerns

Q: Will downsizing my VM impact performance?

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

• 99th percentile (P99) CPU utilization over 14 days
• Peak memory usage
• Network saturation points
• Historical spike patterns

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

Q: What if I don't have memory metrics?

A: Azure Monitor provides memory metrics by default for all VMs. If metrics are missing:

1. Verify Azure Monitor agent is installed
2. Check diagnostics settings are enabled
3. Wait 14 days for sufficient historical data

Q: Can I test without downtime?

A: Yes! Recommended approach:

1. Use create_before_destroy = true in Terraform
2. New VM launches with optimized size
3. Health check passes → traffic shifts to new VM
4. Old VM automatically terminated
5. Typical downtime: <30 seconds during cutover

Performance Issues After Optimization

Symptom: Increased latency or response times

Possible causes:

• Insufficient CPU for peak loads
• Memory pressure
• Network throughput bottleneck

Resolution:

1. Check Azure Monitor CPU, memory, network metrics
2. Compare current vs. previous VM performance
3. If needed, increase VM size by one step
4. Simple rollback: revert Terraform to previous state

Symptom: B-series CPU credit exhaustion

Possible causes:

• Migrated from standard VM to burstable
• CPU usage higher than B-series baseline
• Insufficient CPU credits for workload pattern

Resolution:

1. Check CPU Credits Remaining metric
2. If consistently depleted, migrate to D/F-series
3. Consider B-series with higher baseline (e.g., B4ms → B8ms)

Security Considerations

VM Configuration

Jetscale recommendations preserve:

• Virtual network and subnet associations
• Network Security Groups (NSGs)
• Managed identities
• Key vault references
• Disk encryption settings
• Boot diagnostics configuration

Image Compatibility

• Uses same OS image/SKU as original VM
• Preserves custom images and configurations
• Maintains OS disk and data disk attachments
• No changes to authentication methods

Lifecycle Management

Use Terraform lifecycle rules:

lifecycle {
  create_before_destroy = true
  prevent_destroy       = var.is_production
}

Limitations

Not Currently Supported:

• Virtual Machine Scale Sets (VMSS) optimization
• Spot VM recommendations
• Reserved Instance purchase recommendations
• Auto-shutdown scheduling
• AKS node pool optimization
• Multi-VM orchestration

Roadmap: These features are planned for future releases. Currently, Jetscale focuses on standalone Azure VM optimization where immediate cost savings can be achieved through right-sizing and migration.

API Integration

Jetscale provides API access for programmatic optimization:

# List Azure VM recommendations
GET /api/v1/recommendations?resource_type=azure-vm

# 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 VM optimization?

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

Related Documentation:

• Azure SQL Optimization
• EC2 Optimization