It is almost impossible to identify performance bottlenecks before the program is completely working. Programmers almost never target the right area ahead of time.
— McConnell
Mainly divided into two parts:
- Generating load on the system
- Measuring performance under load
- Involves diagnostics within code
- Is easy to implement and works in almost every environment
- However involves modifying existing codebase for print statements and then having to remove these print statements once finished
- External tools that analyse code at the function or even line-level
- Measures performance by timing code execution given heavy load / input
- Also known as stress-testing software
- Not just restricted to code performance, but how the software performs in the real world (i.e. simulating network requests, etc.)
Suppose have the merge sort algorithm implemented in Python:
- A code profiler would determine that the
merge(list1, list2)operation within themergeSort(list)method is causing the performance bottleneck - A line-profiler would determine that the recursive calls and the array slicing within
merge()takes the longest to execute - Because of this, the result of our profiling tests indicate we must avoid recursive calls to boost performance
- As a result, our merge sort algorithm would need to be re-written to use an iterative approach