Time-sharing systems are used to help balance system resources. Use of these systems often results in irregular gaps between a thread's start and stop time. Also, CPU load will not be uniform and our benchmarking data will not be as useful to us.
There are two suggested strategies to avoid this pitfall:
- Test all code before running benchmarks to ensure things work as they should
- Use JMH to measure only after all threads have started or all threads have stopped
