In this post, I’ll focus on design, code, and deployment inspections for performance.
Inspections are a white-box technique to proactively check against specific criteria.
You can integrate inspections at key stages in your life cycle, such as design, implementation and deployment.
Keys to Effective Inspections
- Know what you’re looking for.
- Use scenarios to illustrate a problem.
- Bound the acceptance criteria with goals and constraints.
The Software Performance Frame is a set of categories that helps you organize and focus on performance issues.
You can use the frame to organize principles, practices, patterns and anti-patterns. The categories are also effective for organizing sets of questions to use during inspections.
By using the categories in the frame, you can chunk up your inspections. The frame is also good for finding low-hanging fruit.
Performance Design Inspections
Performance design inspections focus on the key engineering decisions and strategies. Basically, these are the decisions that have cascading impact and that you don’t want to make up on the fly. For example, your candidate strategies for caching per user and application-wide data, paging records, and exception management would be good to inspect.
Effective performance design inspections include analyzing the deployment and infrastructure, walking the performance frame, and doing a layer-by-layer analysis.
Question-driven inspections are good because they help surface key risks and they encourage curiosity.
While there are underlying principles and patterns that you can consider, you need to temper your choices with prototypes, tests and feedback.
Performance decisions are usually trade-offs with other quality attributes, such as security, extensibility, or maintainability.
Performance Modeling helps you make trade-off decisions by focusing on scenarios, goals and constraints.
Performance Code Inspections
Performance code inspections focus on evaluating coding techniques and design choices. The goal is to identify potential performance and scalability issues before the code is in production.
The key to effective performance code inspections is to use a profiler to localize and find the hot spots. The anti-pattern is blindly trying to optimize the code.
Again, a question-driven technique used in conjunction with measuring is key.
Performance Deployment Inspections
Performance deployment inspections focus on tuning the configuration for your deployment scenario. To do this, you need to have measurements and runtime data to know where to look.
This includes simulating your deployment environment and workload. You also need to know the knobs and switches that influence the runtime behavior.
You also need to be bounded by your quality of service requirements so you know when you’re done. Scenarios help you prioritize.