[PRL] The choice uncertainty principle

[William D Clinger]

Riccardo wrote:
> Some interesting bits, especially at the beginning, bringing into
> question the whole notion of assuming that the hardware underlying
> our software abstractions actually works.

The arbitration problem is a hoary problem that
always comes up when you work with asynchronous
models of concurrent computation, e.g. the actor
model.

As in all science, the truth behind our mathematical
abstractions is a lie.  Our abstractions simplify
reality.  In reality, there are no guarantees.  The
best we can do is to achieve reliability to within
epsilon.  If we can make epsilon small enough, then
the practical benefits of recognizing that epsilon
is positive may not be great enough to justify the
mathematical intractability of keeping epsilon in
our equations.

The trick is to design abstractions for which the
lie works better than the truth.

Will