A Comparison of Sorting Times Between Java 8 and Parallel Colt: An Exploratory Experiment

An exploratory experiment found that sorting arrays of random integers using Java 8's parallel sort required only 50%-70% of the time taken using the parallel sort of the Parallel Colt library. Factors considered responsible for the performance advantage include the use of a dual-pivot quicksort on locally held data at certain phases of execution and work-stealing by threads, a feature of the fork-join framework. The default performance of Parallel Colt's parallel sort was found to degrade dramatically for small array sizes due to unnecessary thread creation.