I agree with everything in this e-mail.  Especially the parts about George
doing a great job.

However I think for this type of development process to be useful, we need
independent software to generate the check values.  The independent program
should emphasize correctness over speed.

If it takes prime95 4 months to test an exponent near 10^7, our New Independent
Slow Program (NISP) could easily spend 12 months on the job.  So to be practical
we might want to carry out the LL computation for only 10^5 or 10^6 iterations
and use those incomplete residues for testing (just like the prime95 self check
does).


Here is something to think about.  Say 50 years from now GIMPS claims to have
found all mersenne primes with exponents below 1 billion.  Some young math
grad student is sitting around contemplating the beauty of mathematics and comes
up with a theory for the distribution of Mersenne Primes, but the GIMPS results
do not match her theory.  What to do?  Doubt the theory or doubt the GIMPS
results?

In order for future generations to trust the GIMPS results, the future will need
to
know enough about the software (versions 1 thru 18 and beyond) and the process
used, to trust the results.  The type of development process you describe will
help
encourage that trust.

Brian Beuning

Ken Kriesel wrote:

>
> I propose the following be considered, for each future version released
> (and for the versions in heavy use currently):
> 1. Code review by qualified volunteers.
> 2. George and such other people as are qualified, determine which exponents
> make good test cases, based on a review of the code.
> 3. Volunteers with some cpu-power run LLtests and double-checks on the test
> cases selected.
> 4. Volunteers run randomly selected exponents LLtests and others run double-
> checks of the same exponents simultaneously.  Random selection may hit
> a test case that's useful but was not identified as a separate case.

________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm

Reply via email to