[QUOTE=c10ck3r;356543]Well, if a factor is found, the ideal thing would be to TELL the person performing the LL and save them a couple thousand GHz-Days that would be wasted, so they can divert resources to a new number that might actually be prime...[/QUOTE]

And, this could actually be done automatically by PrimeNet.

In the case of a "factor found", PrimeNet could respond with "No valid AID" to the LLing worker the next time it checks in. My understanding is if Prime95/mprime sees this message, and the work is not yet underway, it will be removed from the worktodo.txt file so effort is not wasted.

In the case where work is already underway, the client continues the work anyway (I believe; not recently empirically tested). This might be to the worker's wishes -- imagine if someone invested 3,000 GHz Days of effort, but then someone "poached" it and found a factor. Do you think the original assignee would want to lose that work, or rather continue it and get the credit even though the effort was not actually needed? Probably depends on the individual's motivations -- credit or end-goal.

Another dimension to this which has been discussed before: the case where someone poaches someone else's candidate with the same work-type, and reports the result first. It has been suggested that the original assignee should get the credit for the work. This is how GPU72 handles this situation for TFing work.

[QUOTE=Aramis Wyler;356163]Though last I checked you preferred me to work on the lowest numbers in the range up to 79, which would dominate that number. I'm more than happy to work in higher ranges (right 2/5 of the graph) if you think that would help the project more.[/QUOTE]

To put on the table...

The best thing for the project as a whole for the LMH sub-project is to choose "LHM Depth-first", and perhaps increase the depth from 78 to 79 or 80 (or even 81).

Don't "poach".

"Spidy" will automatically bring in the lowest candidates not yet TFed to 79 "bits" as they legitimately become available. Trust us -- almost all of those who take on LLing in that range never complete them, and they eventually are recycled by PrimeNet.

[QUOTE=chalsall;356542]Personally I suspect that's rather unlikely; what would be the benefit to the LLing worker?[/QUOTE]Quicker completions on the smaller exponents. Folks who'll do this are probably less patient than folks who color within the lines. :-)

[quote]But, at the end of the day, everyone's welcome to do whatever they want with their own time, kit and power bills, so long as it doesn't have a negative impact on the project.[/quote]Sure, but there's a discontinuity in the "credit" rewards:

If someone reports a factor 5 minutes before I report my LL on the same exponent, I get LL credit that "sticks".

If someone reports a factor 5 minutes after I report my LL on the same exponent, my LL credit is erased.

It's just a flaw that has potential to give incentive to allocate both LLing and TFing resources where they have no actual benefit for GIMPS.

[QUOTE=cheesehead;356653]If someone reports a factor 5 minutes after I report my LL on the same exponent, my LL credit is erased.[/QUOTE]

Really? IIRC, there is not retroactive deduction of credit in GIMPS. Sure, the LL history is not displayed for factored exponents, but credit (GHz-days) once awarded stays. :unsure:

yep but "History" will forget that "I" ll'ed it. The result is not the same.

[QUOTE=axn;356655]Really? IIRC, there is not retroactive deduction of credit in GIMPS.[/QUOTE]I've seen it happen. Also, there's a specific statement around somewhere (I'll post when I find it) that the purpose was to motivate folks to finish the recommended amount of factoring before doing the LL (plus, implicitly, to take responsibility for verifying that the recommended factoring had been done before starting LL themselves).

The idea was that if a factor of an LLed exponent is found below the recommended factoring levels, that means not enough factoring effort was expended prior to the LL test (and folks had a responsibility to check that before doing an LL, at penalty of possibly losing their LL credit).

This was aimed at users who'd be assigned an exponent that still needed more TF and/or P-1 before LL, but who'd manually intervene to skip the factoring and go directly to the LL, credit for which was their only goal. They didn't care for TF or P-1 credit, but just wanted to rack up the LL credits, and that meant eliminating any chance that LL wouldn't be run.

However, the stated rule did not take into account whether or not the exponent had actually had sufficient factoring done before LL -- it just erased LL credit if any factor was reported later. It seems that this simple rule presumed that folks would not try factoring any already-LLed exponent farther than the recommended limits, so that a post-LL report of any factor implied insufficient pre-LL factoring effort.

nevermind, I figured it out.

I have in my list of ever-done exponents few hundred of "verified", few hundreds of "unverified", some "verified bad residues" and about 10-20 exponent "for which a factor was found after the exponent was LL-ed or DC-ed".

You can check your own LL report to see, there is a high probability you have the last category too.

There was no credit deduction, ever.

I personally found over 100 factors for exponents which were LL-ed and/or DC-ed days before and/or eons before, and many other people work to find factors for "small mersennes" (like under 10M, or 20M expo or so).

Do you mean that if someone finds a factor, the credit of the original LL-er is gone? (this was a rhetorical question, doesn't need an answer).

[QUOTE=cheesehead;356659]I've seen it happen.[/QUOTE]

I vaguely remember, way back when, there used to be two sets of credits, one from Primenet (which was never deducted), and one maintained manually by George, which was sometimes deducted (perhaps for the condition that you mention). Memory being what it is, and not being able to find mersenne.org in the wayback machine, I won't be surprised that I am wrong about all that.

Anywho, I hope that this credit deduction thing is not being done anymore, because that'd be a seriously sucky thing to do in a _volunteer_ computing project.

[QUOTE=cheesehead;356495]You seem to have forgotten that [I]GIMPS's purpose for doing factoring on exponents in ranges about to be assigned for L-L (or higher) is to save the work of doing the L-L test[/I]. That only works [I]when it actually prevents the L-L test from being assigned, not when the L-L has already been assigned[/I].[/QUOTE]

The most efficient way to LL these things is one per core, and it takes something like a year apiece to do them. If 2 months into a LL test I was working on someone TF'd a factor out, I'd be so relieved I might send them a card. They would have saved me 10 months. Having someone stop the wasted work would only be upsetting to a credit whore, not someone who was actually working toward the goal at hand.

The purpose for doing factoring is to cut down useless work, not to cut some arbitrary assignment rate.

What if you were 97% done, and somebody reports a factor. Would you still send him or her the card? LOL