> He means PC's in which CPU speed is not the limiting factor. If your CPU's
> are fast enough, then memory/bus speed could be the limiting factor, and 2.0
> SMP handles this less efficiently than 2.1.
>
> (at least that's what I think he was trying to say :)
I meant that under 2.0, SMP works best if all processors are always
running user-bound tasks. I was about to say "compute-bound" there,
but the real issue is how 2.0 vs 2.1 handle SMP in kernel-mode and
handling interrupts: the 2.1 kernel takes a lot more opportunity to
use multiple processors at the same time. Alan has pointed out that
2.0's big-lock-around-the-kernel actually works pretty well since
2.0 is clever enough to drop it for anything lengthy. 2.1 still has
a lot of advantages, since it uses clever fine-grained locking in
kernel mode, and has pretty radically revamped interrupt handling.
I don't believe that either kernel _can_ do anything to handle apps
which are bound by memory bandwidth, unless we're talking about something
like page-coloring to reduce TLB thrashing.
