While I was testing for the MP kernel panics, I spent a lot of time
watching top, especially during nbmake sessions. I noticed that while
there would be pretty good load sharing between processors, at no time
would one /usr/src/build.sh show that the two processors had less than a
combined 100% idle value. By this I mean that if one CPU was 22% idle,
the other one would be 78% idle, or even higher. This leads me to
conclude that a single parent process can not get more than 100% of
expected CPU time, but that the expected CPU time is perhaps half what
it actually is. If I did a completely separate nbmake, say on a kernel,
I could get both CPUs to nearly 100% usage.

While build.sh was running, the 7300 remained very responsive, as there
was an effective CPU available. However, since build.sh took 21 1/2
hours to build, I think I would have preferred to trade off
responsiveness for a twelve hour build.

Is the scheduler an architectural specific issue, or is this part of
the larger BSD SMP?

thanks,
tim