Profiling is the only way you should make performance optimizations. As you remember "premature optimization is the root of all evil" so don't do this till you don't have a working solution that exhibits performance problems.

See also Intro To Performance

Performance is a catch all term for a lot of things among:

• Time: Fast/Slow
• Memory footprint: Big/Small
• Threads: Many/Few/One
• IO
  • network
  • storage

Best practices

• Two types of analysis
  • instrumentation - you need to filter out methods/packages/libraries that are called a lot of times otherwise this will impact any conclusion
  • sampling - you don't see the actual number of calls and each time duration but an aggregated sampled effort: timePerCall*calls

Tools

• JProfiler has the most intuitive hot spot analysis. Unfortunately is poorly integrated with eclipse and is expensive. It has a 14 days evaluation license else.
  • https://www.ej-technologies.com/resources/jprofiler/help/doc/main/memory.html
• YourKit

Resources

• How to profile https://twitter.com/GeoffreyDeSmet/status/1526464502086131715
  • https://threadreaderapp.com/thread/1526464502086131715.html
• https://stackify.com/java-profilers-3-types/
• http://zvikico.typepad.com/problog/2008/07/jprofiler-your-java-code-could-be-running-faster-in-under-two-hours.html
• https://www.ej-technologies.com/products/jprofiler/ide-eclipse.html
• https://www.ej-technologies.com/download/jprofiler/files
• https://en.wikipedia.org/wiki/Profiling_(computer_programming)
• http://www.baeldung.com/java-profilers