On 11 Sep 00, at 0:31, Robert Deininger wrote:
> I looked at a bit of the database of double-checked exponents:
> ftp://mersenne.org/gimps/lucas_v.zip
>
> I noticed some exponents where the first-time and double-check results
> were submitted by the same person, using the same software, on the same
> machine. If there is a reproducible hardware problem, or one of these
> "bad" multiplications, then this double-check would not catch an error.
If the offset was different then this would be a valid double check.
However I agree it is undesirable for the first test and the double
check for an exponent to be submitted by the same user - though,
where purely automatic methods of obtaining assignments and reporting
results were employed, surely it makes no difference.
Actually this is the reason I'm so keen to definitively eliminate the
very small chance that a systematic hardware problem affecting either
Intel Pentium or Athlon CPU families could be responsible for the
mismatch found during the large exponent QA work. If this were the
case we might have to double-check any exponents which had already
been tested twice using the processor type found to have a problem.
There are also a number of (quite small) exponents where either the
first test or the double check were done using "old" software which
reported much less than 64 bits of residual. I'm steadily working my
way through triple-checking these using systems which are too slow to
be otherwise useful any more. I guess it would probably be worthwhile
running independent triple-checks on those exponents where first test
and double check have been done by the same user - preferably using a
different program as well - but I don't think this task falls into
the "very urgent" category.
> Actually, I kind of doubt that the LL test was done twice. More likely
> one result was accidentally _submitted_ twice, and the friendly primenet
> server happily accepted both.
Hmm. In this case, if the program was George's, the offset (which is
not shown in the lucas_v file) would be the same; hopefully George's
procedure would catch this.
> (We know that the primenet server accepts
> results from people who do NOT have the exponents assigned to them.)
Is this relevant? More than two different people independently
testing any given exponent is a probably unneccessary duplication of
effort, but surely acts to _increase_ confidence in the integrity of
the database rather than reduce it.
>
> Will these seemingly bogus double-checks be weeded out when the next
> synchronization is done?
Hopefully, if there are any bogus double-checks, they will be removed
and any neccessary double-checks added to the list of assignments to
be allocated.
BTW it's a _long_ time since we had a database sychronization ...
Regards
Brian Beesley
_________________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.exu.ilstu.edu/mersenne/faq-mers.txt
