Robert G. Brown
rgb at phy.duke.edu
Wed Feb 19 06:27:11 PST 2003
On Wed, 19 Feb 2003, Simon Hogg wrote:
> Is there a rough rule of thumb which dictates when a program (if ever)
> shows superscaling with number of nodes. Of course, I would not expect
> this to carry on ad infinitum, but does anyone see superscalar behaviour up
> to, a certain number of nodes.
> What would be the conditions for this to occur?
You mean a superlinear speedup of some sort or other? Usually when
distributing the code causes the program to cross some important cache
threshold, e.g. code that swaps on a single CPU not swapping when split
up, code that is highly nonlocal and running out of memory remaining
nonlocal but running strictly out of cache when split up, all while
keeping other bottlenecks (e.g. IPC's) under control.
I don't know if there is any rule, but look for widely separated latency
or bandwidth thresholds that come into play when the code is split.
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
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 at phy.duke.edu
More information about the Beowulf