Call counting

• How call counting works
• Advantages
• Disadvantages
• Performance overhead

How call counting works

Call counting works by instrumenting the application code to record every function call. It measures the number of times a particular method or function is called within the program's execution. Call counting can provide valuable insights into how often specific parts of code are executed, which can be useful for optimizing performance, identifying bottlenecks, and understanding application behavior.

There are various scenarios where using call counting might be more appropriate than CPU sampling or CPU tracing:

1. Identifying high-frequency methods: Call counting excels at identifying which methods are being called most frequently. If you are trying to optimize an application, knowing which methods are called most often can be more valuable than knowing which methods are using the most CPU time.

2. Focused analysis: If you already have an idea of which methods might be problematic and want to confirm your suspicions without the overhead of full tracing or sampling, call counting can provide a focused view on the methods of interest.

3. Minimal overhead requirement: If the system being profiled cannot tolerate the overhead of CPU tracing, and even the lighter overhead of CPU sampling is too much, call counting might be the best option. It involves only incrementing a counter and is usually the least intrusive of the three approaches.

4. Profiling in production environments: Sometimes, you might want to gather performance data from a production environment where it's crucial to minimize the impact on performance. In such cases, the low overhead of call counting makes it an attractive option.

5. Long-running processes: For long-running processes or Java applications with a large number of method calls, the accumulated data from CPU tracing might become overwhelming and difficult to analyze. Call counting can give you a more digestible summary of method call frequencies.

6. Intermittent issues: If your application has intermittent performance issues, CPU sampling might miss short-lived problems that occur between samples. Call counting can give you a clearer picture of method call frequencies over time.

Advantages

• Accurate count of function calls
• Minimal possible performance overhead

Disadvantages

• No timing information
• No stack traces

Performance overhead

Call counting does not capture call stacks or time, but it helps identify which methods are called most frequently. Because it only requires incrementing a counter, its overhead is significantly lower than CPU tracing.

Remember, this technique can still have some overhead, especially if the counter is incremented in very frequently called methods. However, it's less invasive than full tracing and can provide valuable information for performance optimization.

Call counting and the associated overhead can be completely disabled with agent startup option disable_tracing or disable_all.