Performance Testing Workload Modeling: Ramp-Up Duration and

User Load Calculation (Technical Guide)

1. Overview of Workload Modeling in Performance Testing

Workload modeling is a crucial step in performance testing that defines how users interact with a
system over time. The key components include:

• User Load (Concurrency): The number of virtual users (VUs) interacting with the system
simultaneously.

• Ramp-Up Strategy: The rate at which users are introduced into the system.

• Steady-State Duration: The time for which the system is tested at peak load.

• Ramp-Down Strategy: The rate at which users exit the system.

These components ensure that the test accurately simulates real-world user behavior and helps in
identifying system bottlenecks.

2. Factors Influencing Ramp-Up Duration & User Load Calculation

A. Business and SLA Requirements

Workload modeling must align with business goals and Service Level Agreements (SLAs). Key
business requirements include:

1. Expected Peak Load: How many concurrent users does the system need to handle?

2. Transactions Per Second (TPS): Business-defined throughput requirements.

3. Response Time SLA: Acceptable response time for critical transactions.

Example:

For an e-commerce platform expecting 10,000 concurrent users on Black Friday with an SLA of ≤
2s response time, the performance test should simulate:

• Peak load of 10,000 users.

• Average 500-1000 transactions per second (TPS).

• A response time ≤ 2 seconds.

B. Real-World Traffic Analysis

To simulate realistic user load, we analyze historical traffic data from:

1. Monitoring Tools:

1|Pa ge Santhosh Kumar J

 o Google Analytics, New Relic, Dynatrace, AppDynamics provide insights into:

▪ Peak traffic patterns.

▪ Session durations.

▪ User abandonment rates.

2. Web Server & API Logs Analysis:

o Extract peak requests from web server logs:

grep "POST /checkout" access.log | wc -l

o Use log analysis tools like Splunk or ELK Stack to visualize trends.

3. User Distribution Patterns:

o Traffic varies across time zones and business hours:

▪ Enterprise applications: 9 AM - 6 PM peak.

▪ E-commerce: Traffic surges in the evening and weekends.

▪ Streaming services: Peak usage at night.

C. System Capacity & Scaling Considerations

• Fixed Infrastructure: If no auto-scaling is enabled, a slow ramp-up prevents overloading.

• Cloud Auto-Scaling (AWS ASG, Kubernetes HPA):

o The ramp-up should match the auto-scaling threshold (e.g., every 5 minutes).

o Example: If AWS Auto Scaling Group (ASG) scales every 2 minutes, the ramp-up
should introduce 100 users every 2 minutes.

Example: AWS ASG Scaling Thresholds

Metric Scaling Trigger

CPU > 75% Scale up by 2 instances

TPS > 1000 Scale up by 1 instance

Memory > 80% Scale up by 3 instances

3. How to Calculate Ramp-Up Duration?

The ramp-up duration is determined based on:

• Total Users (N): Number of virtual users.

2|Pa ge Santhosh Kumar J

 • Ramp-Up Time (T): Time to reach full load.

• Ramp-Up Rate: The speed at which users are added.

A. Linear Ramp-Up Formula

Ramp-Up Rate=Total Users / Ramp-Up Time (T)

Example:

• 5,000 users over 20 minutes: 5000\20=250 users per minute

o JMeter Configuration: Set Thread Group Ramp-Up = 1200 seconds (20 min).

B. Stepwise Ramp-Up (Batch Injection)

Rather than adding users continuously, they are injected in batches to prevent early failures.

Example: Stepwise Ramp-Up (Batch Injection)

• 1,000 users every 5 minutes until full load is reached.

Time Interval Users Added Total Users

0 - 5 min 1,000 1,000

5 - 10 min 1,000 2,000

10 - 15 min 1,000 3,000

15 - 20 min 1,000 4,000

20 - 25 min 1,000 5,000

JMeter Configuration:

• Use "Ultimate Thread Group" to set ramp-up batches.

C. Arrival Rate-Based Ramp-Up (Little’s Law)

If the system is TPS-driven, we calculate concurrency using:

Concurrent Users=TPS×Response Time

Example Calculation

• Target TPS: 1,000 transactions per second.

• Average Response Time: 2 seconds.

• Required Concurrent Users: 1000×2=2000 concurrent users

• JMeter Configuration:

3|Pa ge Santhosh Kumar J

 o Use "Constant Throughput Timer" to control the TPS rate.

4. How to Determine Total User Load?

There are multiple approaches:

A. Production Traffic Analysis

• Extract peak Concurrent Users from New Relic, Dynatrace, AppDynamics.

• Web Server Logs:

cat access.log | grep "GET /home" | wc -l

B. Capacity Planning & SLA

• Max Load Capacity: If the database supports 2,000 TPS, and avg response time is 3 sec,
then: 2000×3=6000 concurrent users

• SLA-Based Load Testing: Adjust user load to find the maximum sustainable number of
users under the SLA constraints.

5. Performance Test Workload Models

Model Description Use Case

Peak Load Simulates the highest expected user load. SLA validation

Step Load Gradually increases load in steps to find breaking Capacity planning
points.

Spike Sudden large user injection. Flash sales, DDoS testing

Load

Soak Test Sustains load for long periods. Memory leaks, stability
testing

6. Example Workload Scenario

E-Commerce Black Friday Load Test

• Expected Peak Users: 50,000

• TPS Estimate: 4,000 TPS

• Avg Response Time: 3 sec

• Concurrent Users Calculation: 4,000×3=12,000

4|Pa ge Santhosh Kumar J

Ramp-Up Strategy:

• First 10% (5,000 users) in 5 min

• Next 30% (15,000 users) in 15 min

• Final 60% (30,000 users) in 30 min

Steady State: 1 hour Ramp-Down: Gradual drop over 15 minutes.

7. Conclusion

• Linear and stepwise ramp-up are most commonly used.

• Arrival rate-based ramp-up aligns with real-world TPS expectations.

• JMeter can implement these strategies using Thread Groups & Timers.

5|Pa ge Santhosh Kumar J

There are multiple models for calculating ramp-up duration in performance testing. While linear,
stepwise, and arrival rate-based ramp-ups are common, several other methods exist based on
real-world usage patterns, system architecture, and scalability constraints.

Below are other 3 ramp-up models, each with detailed calculations and examples.

4. Spike Ramp-Up (Sudden Traffic Burst)

• All users are injected within a short time (e.g., 1-2 minutes).

• Used for stress testing, DDoS simulation, and flash sales scenarios.

Example Calculation:

• 5000 users injected in 2 minutes: 5000\2=2500 users per minute

Time Interval Users Added Total Users

0 - 1 min 2500 2500

1 - 2 min 2500 5000

JMeter Configuration:

• Use Stepping Thread Group to inject all users quickly.

Use Case:

• Ticket booking, Black Friday sales, breaking news websites.

5. Wave-Based Ramp-Up (Oscillating Load)

• Load fluctuates up and down in waves to simulate real-world traffic fluctuations.

• Useful for handling scaling challenges in cloud environments.

Example Calculation (Waves Every 5 Minutes)

Time Interval Users Added Total Users

0 - 5 min +1000 1000
5 - 10 min -500 500
10 - 15 min +1500 2000
15 - 20 min -1000 1000
20 - 25 min +2500 3500
JMeter Configuration:

• Use Custom Thread Groups to define wave-like traffic.

• Ideal for IoT applications, streaming platforms, and mobile apps.

6|Pa ge Santhosh Kumar J

6. Ramp-Up with Auto-Scaling Considerations

• Designed for AWS Auto Scaling, Kubernetes Horizontal Pod Autoscaler (HPA), Azure VM
Scale Sets.

• Users are added in sync with scaling triggers.

Example Calculation (Scaling Every 5 Minutes)

Time Interval Users Added Total Users Scaling Action

0 - 5 min 1000 1000 1 instance added
5 - 10 min 1500 2500 2 instances added
10 - 15 min 2500 5000 4 instances added
15 - 20 min 5000 10,000 8 instances added
JMeter Configuration:

• Set gradual ramp-up matching cloud scaling intervals.

Use Case:

• Cloud-native microservices that rely on Kubernetes auto-scaling.

Comparison of Ramp-Up Models

Model Best For Ramp-Up Strategy

Linear Ramp-Up General applications Fixed rate increase
Stepwise Ramp-Up Legacy systems Users injected in batches
Arrival Rate-Based Ramp-Up APIs, backend services TPS-driven load
Spike Ramp-Up Flash sales, stress testing Sudden load increase
Wave-Based Ramp-Up Streaming, IoT systems Oscillating traffic patterns
Auto-Scaling Aware Ramp-Up Cloud-based microservices Matches scaling policies

Example Workload Scenario: E-Commerce Black Friday Sale

A retail website expecting 50,000 concurrent users during Black Friday should use a hybrid
ramp-up strategy:

1. Stepwise Ramp-Up for first 10,000 users (to warm up caches, CDNs).

2. Spike Load (20,000 users in 5 min) simulating a flash sale.

3. Wave-Based Load (±5,000 users every 10 min) to mimic real-world purchase behavior.

Test Configuration

Phase Users Ramp-Up Time Load Model

Warm-Up 10,000 10 min Stepwise

7|Pa ge Santhosh Kumar J

Flash Sale 20,000 5 min Spike Load

Steady-State 50,000 1 hour Linear

Load Drop -20,000 15 min Wave-Based

Conclusion

1. Not all performance tests use simple linear ramp-up.

2. Stepwise and wave-based ramp-up models better replicate real-world traffic.

3. Cloud-based tests should align with auto-scaling policies.

4. Arrival rate-based models (TPS-driven) are best for APIs.

5. Spike ramp-up is essential for stress testing and event-driven traffic surges.

8|Pa ge Santhosh Kumar J