Just slightly off topic? (fwd)

[Robert G. Brown]

On Thu, 17 Sep 1998, Bob Glamm wrote:

> This is Amdahl's law.  It basically says that your performance gain
> is limited by the fraction of serial code in your algorithm.  In
> reality it states that you will (usually (*)) never achieve linear speedup.

I agree with this, and deliberately omitted it in my previously sent
analysis.  Well, ok, I forgot it (I got too focused on the
nonlinearities of the parallel part itself -- I >>did<< say "perfectly
parallelizable task:-) However, Amdahl's law frequently contributes a
correction to the LINEAR part of the node scaling -- instead of getting

F(N) = N(1 + ...)

as a series one gets

F(N) = N(A + ...)

where A<1 reflects the serial fraction of code.  (I knew that I included
that A in the final expression I gave for a reason, just forgot to
discuss the reason:-).

Sorry,

   rgb

Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb@phy.duke.edu