From: bill_kempf (williamkempf_at_[hidden])
Date: 2002-03-14 11:26:34
--- In boost_at_y..., "Peter Dimov" <pdimov_at_m...> wrote:
> From: "bill_kempf" <williamkempf_at_h...>
> > How do you disagree with anything I said? Nothing in your
> > description differs from anything I've said, nor does it actually
> > mention any need for any specific mutex scheduling. In a thread
pool
> > it doesn't matter what thread acquires the mutex, and thus
acquires
> > the "job". All that matters is that *some* thread acquires the
mutex
> > and thus the "job". It doesn't even matter if a thread (or set
of
> > threads) is starved in this case, because the jobs are still
being
> > executed in a timely manner. A thread can remain idle
indefinately
> > and not cause any problems with the operation of the thread pool.
>
> Using pooled threads in LIFO manner minimizes page faults.
Ahh... this was the piece I was missing. It's not a correctness
issue, but one of (very platform specific) efficiency. OK, I can buy
this, but the efficiency will matter only to a small few.
Bill Kempf