Re: Benchmarks Game: Thread ring (Was: Large number of tasks slows down my program (using debian) - any fix?)

[Brad Moore <bmoore.ada@gmail.com>]

On Thursday, April 5, 2018 at 11:59:00 PM UTC-6, Jacob Sparre Andersen wrote:
> Brad Moore <bmoore.ada@gmail.com> writes:
> 
> > [...]  I believe the reason why it works as fast as it did, is that an
> > Ada protected object can have one task perform the action of the
> > protected object on behalf of other tasks that are queued on an entry
> > of the protected object.
> 
> > The results comparing this version against other approaches in other
> > languages can be seen here, where the entries are executed on a
> > quad-core machine:
> 
> > https://benchmarksgame-team.pages.debian.net/benchmarksgame/performance/threadring.html
> 
> From the CPU load data, it looks like the Ada solution has four tasks
> running full time, and not just a single one.
> 

Yes, in fact this appears to be the only example that fully loads all the cores. Some of the language examples only involve 1 core, which to me suggests that what is really happening in those cases is something more like a coroutine, which I think is not really a fair comparison.

When I said one task can service the queue on behalf of other tasks, I was thinking that the implementation might be "passing" the baton between the "pseudo"-threads maybe 503 passes at a time, releasing the 503 Ada tasks that are waiting on the entry, then those 503 tasks would queue back up on the entry (not necessarily all 503 at a time), but however many do get in at a time (based on how many are queued on the entry when the entry guard is opened), they would effectively get released at approximately the same time, but still in a serial order. So, I would expect all cores to be fully loaded. I don't know if GNAT actually has the protected object implemented this way, I just know that that is a possible (and I think recommended) implementation.

Cheers,

Brad