I think I can institute a discipline within Keykos, and perhaps other platforms, where calculations are deterministic.
You can detect no noise in a deterministic environment and the environment is thus immune from <a href=../../../Psychology/Category.html>a rather large interpretation of Spectre bugs</a>.
At the same time you can exploit multiprogramming.
Sharing varying resources is out.
The key example is a primitive (trusted) gadget that waits for two pieces of data X and Y produced by two ‘calculations’ and then passes those values to another calculation to work on.
None of the calculations can learn which finished first.
This provides a style of deterministic multi programming.
<p>
I am not yet convinced that this is practical; many useful MP patterns are ruled out.
Many physics applications remain feasible, even convenient.
<a href=http://pascal.eng.uci.edu/projects/sisal/sisaltutorial/00.Contents.html>Sisal</a> fits this plan and Sisal was motivated by easing the development of a class of applications that defined super computing; it can’t be all bad.
<p>
One pattern that is prohibited is shared memory writable by either calculation.
Another is any mutable object that can be invoked by both calculations.
<p>
Keykos claims to support all useful patterns supported by the bare hardware.
This deterministic MP does not.
<p><a href=InKeykos.html>For now</a>