> Can you back your disagreement up with some examples of non-trivial
> programs which are successfully using your model and were unable to
> just as quickly sucessfully use MPI/PVM/etc?
I guess these are considered `trivial' but various rendering
applications scale very well this way. I've done it under linux using
POV-Ray and under IRIX across multiple-Onyx boxes when rendering with
various applications.
Obviously, such programs, which can almost totally disassociate large
parts of themselves (in this case separate frames or regions) can
scale in such ways extremely well...
> I don't believe that that term has any meaning, even on a
> uniprocessor single threaded program. What does it mean? Are the
> registers consistent with memory? Are they consistent only after
> procedure calls? What about store ordering? What about register
> write buffers?
I'm guessing s/he meant that any page on memory was only mapped on
one machine at any time.
> : DIPC makes the programming task easier by allowing the
> : programmer to not worry about invoking the right send/receive
> : calls at the right places in his code. I think this is a great
> : help.
>
> Actual experience has shown the opposite.
Making things easier for the programmer usually doesn't make better
code, applications or help performance. In fact, I would argue the
opposite is true.
-cw
-
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/