thobson / java-stream-benchmarks

Micro benchmarks comparing various stream and non stream implementations of a summing algorithm

java-stream-benchmarks

Micro benchmarks comparing various stream and non stream implementations of a summing algorithm. The benchmark was performed with the JMH benchmark. See this blog post for more information

To run the benchmarks execute mvn clean install && java -server -jar target/benchmarks.jar

To ensure the benchmarks work correctly add the assertions flag java -server -ea -jar target/benchmarks.jar

The stream based implementation are really trivial but they demonstrate the importance of choosing the correct collection implementation. The "traditional" algorithms are more complex but in testing proved to be much faster. For example the standard stream using a LinkedList gave 168 ops/s on my machine whereas an optimised multi threaded for loop gave over 10,000 ops/s.

java-stream-benchmarks's Issues

improved stream API performance

first of all, nice article. Unfortunately there is no possibility to leave a comment, so I have to create an issue on Github ;)

The version 'parallelArrayListStream' could be improved to get comparable performance with the hand-crafted version 'sharedStateThreadPool'.

instead of:

long sum = arrayListValues.parallelStream().mapToLong(Integer::intValue).sum();

do:

long sum = Arrays.stream(arrayValues).asLongStream().parallel().sum();

I think the mapping of integer values to long is the performance killer here.
Greetings