The code given so far is a great option as long as your dataset is small enough to not require scaling out to a cluster. There are some caveats due to the details of how the JDK's Fork/Join engine parallelizes the work, but when done right it will perform very well. The concerns of scaling out, however, have a significant impact on the shape of the computation. While in single-node concurrent computing a major challenge is sharing mutable data, in multi-node distributed computing the major challenge is simply sharing --- of any kind.