18 - Capacity Planning

Right-sizing APIM units, traffic estimation, and scaling strategies

---

🎯 Capacity Fundamentals

What is a "Unit"?

A unit is the base scale factor for APIM. Each unit provides:

Tier	Requests/sec (approx)	Cache	Notes
Developer	~500	10 MB	No SLA, no scale
Basic	~1,000	10 MB	Cannot scale
Standard	~2,500	50 MB	Can add units
Premium	~4,000	5 GB	Zone redundant possible
Basic v2	~5,000	N/A	VNet supported
Standard v2	~10,000	N/A	Higher throughput

⚠️ These are approximations. Actual throughput depends on policies, payload size, and backend latency.

---

📊 Capacity Metric

The Capacity metric is your primary scaling indicator:

Capacity Range	Status	Action
0-40%	Underutilized	Consider scaling in (non-prod)
40-70%	Optimal	Monitor normally
70-80%	Warning	Plan to scale out
80-100%	Critical	Scale out immediately

---

📏 Sizing Methodology

Step 1: Gather Requirements

Metric	How to Measure	Example
Peak RPS	Max requests per second	5,000 RPS
Average RPS	Normal traffic	1,000 RPS
Payload Size	Average request/response	10 KB
Backend Latency	P95 backend response	200 ms
Policy Complexity	Number of policies	5-10 policies
Growth Projection	12-month growth	50%

Step 2: Calculate Base Units

Formula:

Base Units = Peak RPS / Tier Throughput per Unit

Example (Premium tier):

Peak RPS: 10,000
Premium throughput: ~4,000 RPS per unit
Base Units = 10,000 / 4,000 = 2.5 → Round up to 3 units

Step 3: Apply Adjustments

Factor	Multiplier	When to Apply
Complex policies	1.2-1.5x	JWT validation, transforms
Large payloads (>50 KB)	1.3-1.5x	File uploads, large responses
High backend latency (>1s)	1.2x	Slow backends hold connections
Zone redundancy	Minimum 3	Premium tier HA
Growth buffer	1.3-1.5x	12-month projection

Example with adjustments:

Base: 3 units
Complex policies: 3 × 1.3 = 3.9
Zone redundancy: Minimum 3
Growth buffer: 3.9 × 1.3 = 5.07 → Round to 5 units
Final: 5 Premium units

---

📈 Traffic Pattern Analysis

Understand Your Patterns

Pattern	Characteristics	Scaling Strategy
Steady	Consistent load	Fixed units
Cyclical	Predictable peaks	Schedule-based autoscale
Spiky	Unpredictable bursts	Metric-based autoscale + buffer
Growing	Continuous increase	Proactive scaling + monitoring

Traffic Analysis Query

// Analyze traffic patterns
ApiManagementGatewayLogs
| where TimeGenerated > ago(7d)
| summarize Requests=count() by bin(TimeGenerated, 1h)
| summarize 
    Min=min(Requests),
    Avg=avg(Requests),
    P95=percentile(Requests, 95),
    Max=max(Requests)

---

⚙️ Autoscaling Configuration

Metric-Based Autoscaling (Bicep)

resource autoscale 'Microsoft.Insights/autoscalesettings@2022-10-01' = {
  name: 'apim-autoscale'
  location: location
  properties: {
    enabled: true
    targetResourceUri: apim.id
    profiles: [
      {
        name: 'Default'
        capacity: {
          minimum: '3'   // Zone redundancy minimum
          maximum: '10'  // Cost cap
          default: '3'
        }
        rules: [
          // Scale OUT when capacity > 70%
          {
            metricTrigger: {
              metricName: 'Capacity'
              metricResourceUri: apim.id
              timeGrain: 'PT1M'
              statistic: 'Average'
              timeWindow: 'PT5M'
              timeAggregation: 'Average'
              operator: 'GreaterThan'
              threshold: 70
            }
            scaleAction: {
              direction: 'Increase'
              type: 'ChangeCount'
              value: '1'
              cooldown: 'PT10M'
            }
          }
          // Scale IN when capacity < 30%
          {
            metricTrigger: {
              metricName: 'Capacity'
              metricResourceUri: apim.id
              timeGrain: 'PT1M'
              statistic: 'Average'
              timeWindow: 'PT30M'  // Longer window for scale-in
              timeAggregation: 'Average'
              operator: 'LessThan'
              threshold: 30
            }
            scaleAction: {
              direction: 'Decrease'
              type: 'ChangeCount'
              value: '1'
              cooldown: 'PT60M'  // Longer cooldown for scale-in
            }
          }
        ]
      }
    ]
  }
}

Schedule-Based Autoscaling

// Additional profile for known peak hours
profiles: [
  {
    name: 'BusinessHours'
    capacity: {
      minimum: '5'
      maximum: '10'
      default: '5'
    }
    recurrence: {
      frequency: 'Week'
      schedule: {
        timeZone: 'W. Europe Standard Time'
        days: ['Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday']
        hours: [8]
        minutes: [0]
      }
    }
    rules: []  // Same rules as default
  }
  {
    name: 'OffHours'
    capacity: {
      minimum: '3'
      maximum: '5'
      default: '3'
    }
    recurrence: {
      frequency: 'Week'
      schedule: {
        timeZone: 'W. Europe Standard Time'
        days: ['Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday']
        hours: [20]
        minutes: [0]
      }
    }
    rules: []
  }
]

---

💰 Cost-Capacity Tradeoffs

Premium Tier Pricing (Approx)

Configuration	Units	Est. Monthly Cost	Use Case
Minimum HA	2	~$5,600	Small production
Zone Redundant	3	~$8,400	Standard production
High Availability	4	~$11,200	Important workloads
High Throughput	6	~$16,800	High-traffic APIs
Enterprise	10	~$28,000	Large scale

Optimization Strategies

Strategy	Potential Savings	Implementation
Right-size for actual traffic	20-40%	Monitor and adjust
Autoscale down off-hours	15-25%	Schedule-based scaling
Caching	10-30%	Reduce backend calls
Reserved instances	30-50%	Commit to 1-3 years
Use v2 tiers	Varies	Evaluate Basic/Standard v2

---

📊 Load Testing

Azure Load Testing Integration

# load-test.yaml
version: v0.1
testName: APIM Load Test
testPlan: apim-load-test.jmx
description: Validate APIM capacity under load
engineInstances: 5
failureCriteria:
  - avg(response_time_ms) > 500
  - percentage(error) > 5

Key Metrics to Capture

Metric	Target	Concern Threshold
P50 Response Time	< 100ms	> 200ms
P95 Response Time	< 500ms	> 1000ms
P99 Response Time	< 1000ms	> 2000ms
Error Rate	< 0.1%	> 1%
Throughput	Target RPS	< 80% target
Capacity	< 70%	> 80%

Pre-Production Load Test Checklist

• Test with production-like payload sizes
• Test with production-like backend latency
• Run sustained load (30+ minutes)
• Test burst scenarios (3x normal)
• Verify autoscaling triggers
• Capture baseline metrics
• Test during scale-out operation

---

📋 Capacity Planning Checklist

Initial Sizing

• Peak RPS estimated
• Payload sizes documented
• Backend latencies measured
• Policy complexity assessed
• Growth projection calculated
• Base units calculated
• Adjustments applied
• Zone redundancy considered

Ongoing Management

• Capacity alerts configured
• Autoscaling rules defined
• Monthly capacity review scheduled
• Quarterly right-sizing review
• Load testing before major releases
• Budget alerts configured

---

🔢 Quick Reference Tables

Units by Traffic

Daily Requests	Avg RPS	Recommended Units (Premium)
1M	~12	2-3
5M	~58	3
10M	~116	3-4
50M	~580	4-5
100M	~1,160	5-6
500M	~5,800	8-10

Scaling Times

Action	Approximate Time
Add 1 unit	15-30 minutes
Remove 1 unit	15-30 minutes
Update from 2→5 units	30-45 minutes

---

🔗 Related Documents

Document	Description
09-Cost-Optimization	Cost strategies
02-Reliability	HA requirements
12-Tradeoffs	Capacity vs cost

---

Back to: README - Main documentation index