Yes, Richard's results show a wild variance that can't easily be
explained. And yes, that means that until there is an explanation for
the variance, the results are not a suitable basis for any kernel
modifications, or even a discussion of such modifications.
But part of Richard's reasons for writing to the list, I suspect, is
to say: "look, here's my code, here's my results, i've thought about
this and i don't understand it. do any of you kernel folk have any
insights ?".
If you (or I or anybody else) doesn't understand it either, then we
should just say say so (or say nothing) and then be quiet. Telling
Richard over and over that his benchmark is wrong isn't doing anything
to either get him to completely rethink his design or answer the
question of where the variance comes from. If the variance comes from
a bad design in his test, it ought to be possible to state what is
wrong. People have tried to do that, and Richard has, to my mind,
offered at least a reasonable defense of what he's trying to do. If
you can't explain his error to him so that he can see it, then just
stop trying. I have to say that nobody so far, not Linus, not David
Miller, not Larry McVoy, have explained his design error in a way that
makes it clear to me, at least. I don't know exactly *what* he's
measuring, but I think he stands a good chance of measuring
*something* interesting.
On the other hand, if the variance comes from something real, then we
should all have some interest in what it is.
--p
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/