Hi Jacob,
On Wed, 9 Feb 2011, Jacob Keller wrote:
> Regarding the last point, does anybody have a good response to the
> Moore's law conundrum that some programs which will take, say, ten
> years to run now will take only ~1 year to run 8 years from now,
> making it futile to run the program now? Maybe it is never worth it to
> run such processes, assuming Moore's law will continue?
This assumes that the process stays running on the same machine. If you
checkpoint it and migrate it to faster machine(s) as they become available,
you may finish earlier. This depends on how processing throughput for the
particular problem evolves between now and now + 8 years. If you choose to
re-code/re-optimise to take full advantage of newer machines, the time
involved in doing that must be factored in as well: that could be a
fully-fledged research project in its own right.
Regards,
Peter.
--
Peter Keller Tel.: +44 (0)1223 353033
Global Phasing Ltd., Fax.: +44 (0)1223 366889
Sheraton House,
Castle Park,
Cambridge CB3 0AX
United Kingdom
|