On Wednesday 13 September 2006 19:04, Chris Caldwell wrote: > > > It seems I've been DOUBLE CHECKING literally _all year_. How can I > > > get out of that 'useful' task and back into the job of high stakes > > > factoring?
On the "PrimeNet" menu, check "factoring assignments" and uncheck everything else. > > I would like to add, that as a mathematician, I value the double checks as > much as the new Mersenne finds. We can not say if the largest record is > the 44th Mersenne until the double checks are complete. Single checks are > not enough because of the 2% (or so) error rate of first checks. Second > checks are much more accurate because it is not the prime/composite final > answers that are compared, but a residue from the calculation which are > **very** unlikely to match after an error. (And if they do not match you > have the valuable triple check which helps check the error rate of the > first check...) Thanks, Chris. > > There out to be a double/triple checkers hall of fame for the those who > checked the most (maybe there is?) Maybe I'm "guilty as charged". On a slow system (currently an old laptop powered by a 366 MHz Celeron) I'm systematically triple checking small exponents, at present triple checking is complete to exponent 1,320,000. If you look at the lucas_v database you'll find that the huge majority of the exponents below 1,320,000 have one entry for username "beejaybee". Mostly the point of this is to ensure that current versions of the software are returning residues consistent with the old versions current when these exponents were originally tested & double-checked. No problems so far. I'm also triple checking selected exponents where both the logged L-L tests have the same username. I have the assistance of one volunteer in this task and, at the moment, we're just about keeping up with double checking (most of the exponents we're working on are in the 16 million range), though there are a large number of exponents still requiring verification. In fact the proportion is increasing; worryingly there are a lot of exponents in the "first test" range with default user & computer names - not just random reassignments to the same user or group - I guess this is due to the proliferation of systems with dual core CPUs. Anyone who would like to join this effort, please contact me directly. Meanwhile I would strongly suggest that dual CPU users set up two assignment streams rather than running the L-L and double check in parallel, which seems to me to be unreasonably risky, not to mention the possible risk of users jumping up the ranking table by fraudulently submitting "double checks" that haven't actually been run. Regards Brian Beesley _______________________________________________ Prime mailing list [email protected] http://hogranch.com/mailman/listinfo/prime
