현실적인 작업 부하 설계

Welcome to "Crafting Realistic Workloads"! In performance testing, a workload model defines how virtual users interact with your system during a test.

It's crucial because a realistic model ensures your test results accurately predict how your system will perform under real-world conditions, helping you find actual bottlenecks.

Start by identifying your user personas. Who uses your application? (e.g., guest, registered user, administrator). Each persona might have different behaviors.

Then, map out their typical business processes or "user journeys." What steps do they take? (e.g., "Guest browses products," "Registered user adds to cart and checks out").

A transaction mix describes the proportion of different actions users perform. For example, 70% of users might browse, 20% search, and only 10% make a purchase.

This mix should reflect real-world usage patterns, often derived from analytics data or business requirements, to accurately simulate system load.

Think time (or pacing) is the pause a real user takes between actions. Without it, virtual users send requests as fast as possible, which is unrealistic and can overwhelm the system.

Adding realistic delays simulates human interaction, making your load test more accurate and preventing premature bottlenecks.

A load test typically has three phases: ramp-up, steady-state, and ramp-down.

• Ramp-up: Gradually increases the number of virtual users.
• Steady-state: Maintains a constant peak load for a duration.
• Ramp-down: Gradually decreases the number of virtual users.

These phases simulate how user load changes over time in the real world.

For a truly realistic workload, each virtual user should use unique, varied data. This prevents caching effects from skewing results and simulates diverse user inputs.

Instead of logging in with "user1" repeatedly, simulate multiple users like "user1", "user2", "user3", each with their own unique data set for actions.

Where do you find the data to build your workload model?

• Web Analytics: Tools like Google Analytics provide user behavior patterns.
• Server Logs: Access logs reveal request frequencies and popular endpoints.
• Business Intelligence: Reports on sales, user activity, etc.
• Product Owners/Managers: Direct insights into expected user journeys and peak usage.

In JMeter, the Throughput Controller is excellent for defining a transaction mix. It allows you to specify a percentage of executions for its child elements.

For example, you can have one controller for "Browse Products" with 70% throughput and another for "Checkout" with 10% throughput, reflecting your desired mix.

k6's scenarios block is powerful for modeling diverse user behaviors. You can define multiple scenarios, each with its own executor, VUs, and duration, to simulate complex workloads.

Here's an example simulating different user types with distinct load patterns:

Which of the following are essential components when designing a realistic workload model for performance testing? (Select all that apply)

In this lesson, we explored the critical elements of crafting realistic workload models for performance testing.

• We covered defining user personas, business processes, and transaction mix.
• We emphasized the importance of pacing, concurrency patterns, and dynamic data.
• Finally, we looked at how to implement these models using tools like JMeter and k6.

A well-designed workload is the foundation for meaningful performance test results!