> Hi,
>
> I have just check to world test status, assigment report. I was horrified
> to see than some individula LL-test have still years (no joke) of computation
> in front of them!
Yeah, I noticed this too. No-one seems to be bothered particularly.
Sometimes this is caused by the system being unusually busy at the
time they checked in their progress, this can cause the system to be
temporarily very slow so far as Prime95/NTPrime is concerned, this
makes the "time to run" estimate much longer than it should be.
> Maybe, It would be wise to communicate with these people
> and:
>
> 1-Check if they are still working on their problem.
I've suggested reassigning the exponent if the user doesn't check in
a completion date for a long time - e.g. 4 months - this will catch
those who have given up, even if their completion date is way ahead.
Also this would catch those users still using 15.x, which doesn't
have the same automatic mechanism for intermediate check-ins.
> 2-Suggest them to upgrade their program.
Uh? It's quite a few versions ago that the last significant
improvement to speed was made. I don't think this suggestion would be
helpful...
> 3-Ask them to go on a faster machine.
Again, if they had a faster machine, maybe they'd already be using
it.
> 4-Redirect them to a more productive task (i.e. Double-Checking small number).
This is hard. Double-checking really needs to be done on different
hardware & using a different program. The problem is that none of the
alternative programs are anything like as efficient as Prime95. And
there are very, very few exponents which have been tested using
something else, but not with Prime95/NTPrime/mprime on an Intel CPU
based system.
I think trial factoring is far enough ahead that encouraging slow
machines to do that isn't going to be helpful. However, slow machines
can still run ECM factoring usefully. I think it would be better to
encourage them to try ECM.
If you're *really* bothered about the exponents with estimated
completion dates way, way ahead, there's nothing to stop you running
the tests yourself (without being assigned them by PrimeNet) &
manually checking in the results. However, if you do this, you're not
eligible for the cash prize, even if you do discover a prime.
If you use a non-PC system, then you won't even disturb the original
user, it's just that his/her test will change to a double-check.
> By the way, I think when the server for non-PC base client will be in
> place. It
> will be primordial to do some publicity. Otherwise, GIMPS will be limited by
> the lack of computer power available for double-checking.
Agreed. But the *real* problem is that there are lots more PCs
around (Intel CPU, running Windows or Linux) than anything else. Also
the fastest alternative program running on a Sun Ultra 10 (64-bit CPU
at 300 MHz) is hardly any quicker than a P90 running Prime95.
Also, the "carrot" of the ca$h prize is only available for running LL
tests. The chance of finding a prime by restesting a number which has
already been tested is not very great (to put it mildly). Running
only factoring (either trial or ECM) gives you no chance at all.
Anyone any ideas for motivating people to try double-checking, or ECM
factoring?
I'd *suggest*:
(a) using a multiplier for CPU time accredited for independent double
checks - so people would rise up the rankings faster if they did
them;
(b) an annual prize for the numerically largest factor found during
each calendar year - though I don't know where the money would come
from.
Regards
Brian Beesley
p.s. if anyone has access to a Sun Sparc or Ultra (ASK THE SYSTEM
MANAGER FIRST - EVEN IF YOU ALREADY HAVE A USER ACCOUNT - THE CPU
TIME USED MAY ALARM HIM/HER IF NOT FOREWARNED!) and can't (or can't
be bothered to) compile MacLucasUNIX from the source, please mail me
& I'll send you a "ready-to-run" binary.
