Reusing the map between benchmark call spoils the benchmark. The first run needs to allocate data structures that might remain after a call to clear(). Allocate

Reusing the map between benchmark call spoils the benchmark. The first run needs to allocate data structures that might remain after a call to clear(). Allocate a fresh one for each run. And use JMH, run some iterations. You might be surprised at the variability.