accumulate locally and mutatively but combine concurrently (a use case)

[John Rose]

On May 10, 2013, at 4:59 PM, Doug Lea <dl at cs.oswego.edu> wrote:

> Parallel reduction/accumulation is of course an old and well-studied
> area, so there are many known good ways and even more known bad ways
> to handle particular cases. But for collecting into hash tables,
> we're pretty sure that using CHM in CONCURRENT mode is the best
> available choice.

I accept that, with pleasure, assuming there are more table keys than processors.

The case I'm asking about is the keyless case, which seems to benefit from buffering before inter-processor communication.  In this case, there is no need to move the data once it has been buffered.

— John