Java 并行流处理时间不匹配

Question

On my windows machine, I have 6 cores.

I create a stream of 24 elements, and execute an operation on each element using map operation in Stream API. I make use of parallelism:

myCollection.parallelStream()
                    .map(element -> someTimeConsumingOperation(element)
                    .collect(toList());

someTimeConsumingOperation simulates long operation using:

Thread.sleep(2000);

myCollection has 24 elements. I understand that streams use default ForkJoinPool executor to run operations in parallel, which in turn takes into account number of cores to determine how many threads to have in the pool.

As there are 6 cores, so I assume max 6 threads are created. If that is the case, and every operation takes almost 2 seconds, entire stream processing must take around 8 seconds(each thread would execute 4 operations as we have 24 elements to process, hence 4*2 seconds).

But I observe that processing takes just 4 seconds.

Answer 1

Your 6-core Intel CPU has HyperThreading enabled. With HT, the operating system and applications see each CPU core as 2 separate "logical cores" or "threads" (the nomenclature varies). This means ForkJoinPool will use 12 threads instead of 6.

The "logical cores" share some resources between each other, so running 12 tasks in parallel does not give you twice the throughput as running 6 tasks. But a program that only calls sleep() won't know the difference.