On 11 Mar 2001, at 7:55, Steve wrote:
> >> Another solution that will work: Have as default a 7 day check in period
> >> at most and only a grace period
> >> of 7 days (not 60). Let the user set the check in period to a higher
> >> value only via the expert menu and
> >> after results have been checked in. That way abandoned exponents get
> >> released in 14 instead of 80 days.
> >
> >That idea sounds the least painful of all that have been discussed so far
> >(to me at least), and since the discussion of what people want in a new
> >version of Prime95 has also been floating arround... this sounds like a
> good
> >suggestion to be submitted. It would not only take care of a problem, but
> >would also not be so harsh to those who own slower machines. A win-win
> >situation from those points of view. Great idea, Martijn!!
>
> I agree this is a good idea, although the 7-day grace period may be a little
> drastic. But even reducing that just from 60 to 30 days (along with a 7-day
> default check in) would release abandoned exponents in 37 days instead of
> 88. This would recycle them more than twice as fast, greatly enhancing the
> odds of someone eventually getting the assignment who will actually finish
> it.
The downside is that there would probably be a great increase in the
number of assignments which are still running but don't complete
before the expiry date. Clearly there is a balance to be struck
somewhere, but 7 days seems to me to be _ludicrously_ short.
In fact, as assignments take progressively longer to run, the "grace
period" should be extending, rather than contracting.
We should also bear in mind the very valuable contributions made by
those people who do not have permanent (or near-permanent) network
connections, and those people who are using clients without the
PrimeNet communication protocol. Requirement to check in frequently
is off-putting to these people. (Some would put it a lot stronger
than that!) I don't think we want to risk driving these people out of
the project.
> As others have mentioned, the problem is NOT slow machines but
> rather abandoned exponents, which has nothing to do with machine speeds.
I fail to see how reducing the check-in interval would have any
impact on the "problem". Those people who are checking in every 28
days aren't running into the 60-day expiry deadline.
The 60 day expiry value is a server parameter, not a client
parameter. In any case, as I explained above, I think that a drastic
reduction in the value would be dangerous.
Might I suggest a couple of alternative approaches. Both of these
would require the identification of exponents which are "seriously
lagging" - perhaps the 100 smallest outstanding LL and the 100
smallest outstanding DC assignments.
(1) Removing these assignments from PrimeNet and managing them
seperately. Anyone who is prepared to make special arrangements to
acquire these assignments is unlikely to default by reason of lack of
commitment.
(2) Alternatively, awarding double PrimeNet CPU time credit for the
completion of these assignments. The downside to this is that, as
well as requiring changes to the server software, recycled "small"
exponents would have to be released at random times of the day, to
prevent them being systematically "grabbed" by a few users.
Regards
Brian Beesley
_________________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
