On Tue, Jun 29, 1999 at 12:12:28PM +0200 (OK, late reply, it suddenly
struck me that I hadn't replied...), Sturle Sunde wrote:
>number which is tested already, you climb by pushing someone else down.
That isn't very likely to happen, is it? Am I the only one who doesn't
trial-factor random LL tested exponents? :-)
>If I don't factor far enough, that will eventualy happen to me.
Hmmmm, perhaps...
(Of course, factoring is a good idea in general; it saves you from
wasting times on LL tests.)
>Therefore I think that Georges
>formula, just counting LL-results for every Mersenne without a factor
>in the database and give credit to the people who tested those numbers,
>is a beautiful solution.
But what if a person had a load (say any number for the discussion, 1000
might be a bit extreme, but still _possible_) of 486s only, and didn't
want them to LL test becuase that takes _ages_? (As we've discussed
earlier, some Dells have problems with flickering during LL tests, which
I've experienced myself the last two weeks.)
(Then again, it saves some problems -- people with P6-class CPUs (that have
almost twice as much `factoring speed' as `LL speed' per cycle
(in P90 CPU year)) won't be tempted to do factoring only, and run up
the ranks :-))
As long as you do get assigned normal, first-time LL tests (double-checks
are already trial-factored with only a marginal chance of a factor missed,
so you get an unfair advantage -- no factoring has to be done), George's
setup is perfect. When things get a bit more complex, IPS is better, IMHO.
/* Steinar */
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
