Poaching - Page 3

eepiccolo · 2003-01-20, 18:23

I figured I'd look at the benchmark page, and get some actual numbers for specific time frames.
For all the following, I entered that the computer would be on 16hrs out of the day, figuring that 16hrs would probably be the least amount of time that Prime95 would get if the computer was left on 24hrs a day.
For a first check, I was generous and chose 20,000,000, I found that a 633Mhz Celeron can do it in six months, a 33Mhz Celeron can do it in nine months, and a 233Mhz Celeron can do it in a year.
For a double check, I chose 10,000,000, and I found a 233Mhz Celeron can do it in three months, a P133 can do it in six months, and P90 can do it in nine months, and a P66 can do it in a year.

Personally, I'd be for restricting DC to six months, and first times to a year. If people want to use slower machines than the ones I listed, or if they don't have a lot of CPU time devoted to Prime95 in a day, they can TF with those machines. A time limit on the ten-million digit Mersennes I think is unnecessary at this point since they don't hold up any "Benchmarks."

eepiccolo · 2003-01-20, 18:59

I forgot to mention in my post that the time limit would apply not only when the exponent is assigned, but also if the current progress is checked in, and the expected completion date has slipped behind the time limit. This may seem a bit harsh, but it probably is necessary. We need some way to make sure the user understands what it means when he/she enters a number for how many hours a day Prime95 will get CPU time.

Also, to prevent poaching, the server could do the following. When it assigns exponents, it can also make a passcode corresponding to each exponent. This passcode would go in the worktodo.ini file with the exponent, and probably would also show up in the results.txt file. That way, you need the passcode to submit the exponent. Sure, someone could go ahead and poach work, but they wouldn't be able to submit it to GIMPS and get credit without the passcode. However, I think this should only be established if we have some sort of time limit as I and others have suggested, so that we don't get stuck waiting on a ten year DC.

ET_ · 2003-01-20, 20:07

Quote:

I forgot to mention in my post that the time limit would apply not only when the exponent is assigned, but also if the current progress is checked in, and the expected completion date has slipped behind the time limit.

I don't agree with you.
I have about 10 PCs working on GIMPS in a travel agency. The agency is working Mo/Fr , 9:00-18:00 or, approximately 40/45 hours a week.
Why should I turn off a 450 hrs/week testing network (that is 2,5 PCs working 24/7) just because each PC alone can't reach the minimum amount of work?

Luigi

QuintLeo · 2003-01-20, 20:25

I still think a fixed scale is a bad idea.

Longer exponents should have longer time limits.

As far as my comment about trial factoring - well, I do sometimes forget that some machines running GIMPS don't run 24/7 - I was looking at my K5-75 factoring-only box that manages to factor an exponent in a couple months or thereabouts, at the currently available "factoring available" size.

15million digit "being factored" exponents aren't holding anything up at the moment, and don't seem likely to be a holdup for a while. Plus a lot of trial exponents *get* factored in the trial factoring stage, and then don't need to be LL checked at all....

In any event, if there is a time limit on a factoring job, eventually there *will* be folks excluded from participation. Lotta 386 machines still around, and many of them are the only computer someone owns - exclude that person *now*, and they likely won't come back when they get that brand new Hammer 4000+ box next year....

Prime95 · 2003-01-20, 21:34

Quote:

Originally Posted by ET_

I have about 10 PCs working on GIMPS in a travel agency. The agency is working Mo/Fr , 9:00-18:00 or, approximately 40/45 hours a week. Why should I turn off a 450 hrs/week testing network (that is 2,5 PCs working 24/7) just because each PC alone can't reach the minimum amount of work?

Assuming these are 400 MHz Celerons working 1/4 time, they can double-check an exponent around 10,000,000 in less than 6 months. Considering our double-checking milestone is at just 7,000,000 I don't see any problems at all. We just need to improve the server so that you don't get an exponent near 7,000,000 because someone might be tempted to poach it before you complete.

nomadicus · 2003-01-20, 23:17

I am still a bit new to all this and never thought of poaching as being a real possibility. How much is the $100,000 prize a factor?

To a certain extent poaching is already declared a possibility. From c:\prime95\License.txt I read
"7. If you were to find a 10,000,000 digit prime . . . Your chance of success is roughly 1 in 250,000. Furthermore, someone may find a 10,000,000 digit prime before GIMPS does."

The help file prime95.chm says nothing on poaching. Should it?

Quote:

If that were the case, I could check out 10,000 exponents and just sit on them and eventually get credit for all of them... (Yes, it is possible to check out 10,000 exponents with just a little work!)

Ouch. How 'bout only being able to check out only so many based on . . . oh never mind, you guys already discussed that line of thinking.

Lastly, the readme.txt file talks about a manual method, but I don't get it. How can an M be assigned without GIMPS knowing about it? I have to give it a username and password. I mean if I stash a "Test=1234567" in my worktodo.ini can't GIMPS validate if it gave it to someone else and tell me to buzz off -- someone else has it?

I hope I'm not asking too many questions. I won't even ask about what to do with the PII 266MHz machines . . .

binarydigits · 2003-01-21, 01:36

Nomadicus: the exponents being poached are in the 2 million digit range and thus not eligible for the $100000 prize, so it is not a factor at all. The poaching usually seems to come up whenever we are approaching a milestone and some stragglers are holding it up. Then someone (or ones) take it upon themselves to 'clean up' the trailing exponents. That's why we have been discussing this as if preventing stragglers will prevent poaching; it undoubtedly would do away with most of it.

Yes, you can type any exponent you want into worktodo. The GIMPS server will give you an error message if you check in an exponent which is assigned to someone else, but the client will continue working on an exponent it has already started, no matter what. (If not yet started, it will be deleted from the worktodo.) But once you complete it and check in the result, you will get the credit and the person who has the assignment will start getting messages "exponent already tested". This has happened to me on numerous occasions where I picked up a recycled exponent and the person who had it before me (and let it lapse) then finished it before I did. If my machine has already started on it, I have to manually delete it from the worktodo. If it is too far along I just let it go ahead and finish as an early doublecheck. But now I'm drifting off topic (not to mention ranting), sorry.

Even though you didn't ask about your 266MHz PII, I will tell you that I have a 233MHz Pentium Pro doing trial factoring. Takes it 19 days to complete an assignment in the current range, but I consider it far from useless. I also have a couple of 300MHz PIIs doing doublechecks, taking about 40 days for 8.9M exponents. A friend of my nephew's is even running DCs on a 120MHz Pentium (his firewall box), takes about 4 months. So your 266MHz machine still has plenty of use... but I wouldn't recommend 1st time LL tests :-)

2003-01-21, 02:52

It is as it is, PERIOD.....

2003-01-21, 03:02

If someone, manually tests an exponent, that is reserved by someone else, they could then use LLR to announce primality before the reserved slowpoke.

Claiming the full prize!

This is the only problem that should be weighed, for time on an exponent.
In other words If I can test an exponent with a p4, then test it with LLR too, quicker than the reserved PC, then there is probable cause to unreserve it.

It should be a mathematical equation, with a variable time limit.

Prime95 · 2003-01-21, 04:34

Quote:

Originally Posted by TTn

If someone, manually tests an exponent, that is reserved by someone else, they could then use LLR to announce primality before the reserved slowpoke.

Claiming the full prize!

It is a violation of the source code copyright to use LLR to prove a Mersenne number prime and claim the EFF prize.

In theory, you could find a 10 million digit prime and not report it to the primenet server. Then you could write your own LL test program and file a claim for the prize. Then all you need to do is convince EFF that you wrote an LL testing program when there was a publicly available one, tested one exponent and got lucky. Methinks it would be easier to thank your lucky stars and let GIMPS help you claim half the prize :)

cheesehead · 2003-01-21, 06:42

Quote:

Originally Posted by QuintLeo

Lotta 386 machines still around, and many of them are the only computer someone owns - exclude that person *now*, and they likely won't come back when they get that brand new Hammer 4000+ box next year....

Exactly.

I have a "slow" system which, despite its "slow"ness, never, ever held up any milestone while LLing. I kept careful track of GIMPS progress, and made sure my system's work was always on schedule to report results sufficiently ahead of milestones.

Nevertheless, it seems to have triggered one particular poacher's ire because he kept coming back and coming back to poach my assignments. (Not necessarily that he targeted my userid -- maybe it's just that his algorithm for selecting poaching targets happened to keep selecting mine.)

Because at first I did not understand what was happening when my first-time LL assignments suddenly changed into DC assignments, I did not gather the evidence to conclusively demonstrate that six of the first eight occurrences were due to poaching by "Mr. X". (He _did_ LL the same exponents I was assigned -- I just can't prove the timing.) Once I did realize what was happening, I gathered enough data to show that "Mr. X" very probably poached two more of my LL assignments, and definitely, unmistakably poached my final LL assignment under the userid I was then using. In every case, my system was on schedule to complete the assignment without holding up any milestone.

I have changed userids and done some other things to avoid "Mr. X"'s (should I name his name here?) unwelcome attentions.

But note that despite the fact that all my assignments were on track to be completed ahead of milestones, "Mr. X" apparently concluded that my system was some sort of straggler.

How will any of the proposed changes to PrimeNet keep non-stragglers from appearing to be stragglers? The existing system does not succeed in that, apparently.

2003-01-21, 02:52	#30
TTn 22773₈ Posts	poaching, smoaching!!! It is as it is, PERIOD.....

2003-01-21, 03:02	#31
TTn 3·2,731 Posts	poaching loophole If someone, manually tests an exponent, that is reserved by someone else, they could then use LLR to announce primality before the reserved slowpoke. Claiming the full prize! This is the only problem that should be weighed, for time on an exponent. In other words If I can test an exponent with a p4, then test it with LLR too, quicker than the reserved PC, then there is probable cause to unreserve it. It should be a mathematical equation, with a variable time limit.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Poaching	blip	Data	8	2016-01-30 01:59
Poaching	davieddy	Lounge	6	2010-10-16 12:31
Poaching and v5	PrimeCruncher	PrimeNet	6	2004-04-05 19:17
Officially poaching very old exponents	Prime95	Data	17	2003-11-13 02:13
New fashion poaching (???)	lycorn	Lounge	6	2003-01-31 08:33

2003-01-20, 18:23	#23
eepiccolo Dec 2002 Frederick County, MD 101110010₂ Posts	I figured I'd look at the benchmark page, and get some actual numbers for specific time frames. For all the following, I entered that the computer would be on 16hrs out of the day, figuring that 16hrs would probably be the least amount of time that Prime95 would get if the computer was left on 24hrs a day. For a first check, I was generous and chose 20,000,000, I found that a 633Mhz Celeron can do it in six months, a 33Mhz Celeron can do it in nine months, and a 233Mhz Celeron can do it in a year. For a double check, I chose 10,000,000, and I found a 233Mhz Celeron can do it in three months, a P133 can do it in six months, and P90 can do it in nine months, and a P66 can do it in a year. Personally, I'd be for restricting DC to six months, and first times to a year. If people want to use slower machines than the ones I listed, or if they don't have a lot of CPU time devoted to Prime95 in a day, they can TF with those machines. A time limit on the ten-million digit Mersennes I think is unnecessary at this point since they don't hold up any "Benchmarks."

2003-01-20, 18:59	#24
eepiccolo Dec 2002 Frederick County, MD 172₁₆ Posts	I forgot to mention in my post that the time limit would apply not only when the exponent is assigned, but also if the current progress is checked in, and the expected completion date has slipped behind the time limit. This may seem a bit harsh, but it probably is necessary. We need some way to make sure the user understands what it means when he/she enters a number for how many hours a day Prime95 will get CPU time. Also, to prevent poaching, the server could do the following. When it assigns exponents, it can also make a passcode corresponding to each exponent. This passcode would go in the worktodo.ini file with the exponent, and probably would also show up in the results.txt file. That way, you need the passcode to submit the exponent. Sure, someone could go ahead and poach work, but they wouldn't be able to submit it to GIMPS and get credit without the passcode. However, I think this should only be established if we have some sort of time limit as I and others have suggested, so that we don't get stuck waiting on a ten year DC.

2003-01-20, 20:25	#26
QuintLeo Oct 2002 Lost in the hills of Iowa 2⁶·7 Posts	I still think a fixed scale is a bad idea. Longer exponents should have longer time limits. As far as my comment about trial factoring - well, I do sometimes forget that some machines running GIMPS don't run 24/7 - I was looking at my K5-75 factoring-only box that manages to factor an exponent in a couple months or thereabouts, at the currently available "factoring available" size. 15million digit "being factored" exponents aren't holding anything up at the moment, and don't seem likely to be a holdup for a while. Plus a lot of trial exponents get factored in the trial factoring stage, and then don't need to be LL checked at all.... In any event, if there is a time limit on a factoring job, eventually there will be folks excluded from participation. Lotta 386 machines still around, and many of them are the only computer someone owns - exclude that person now, and they likely won't come back when they get that brand new Hammer 4000+ box next year....

2003-01-21, 01:36	#29
binarydigits Aug 2002 64₈ Posts	Nomadicus: the exponents being poached are in the 2 million digit range and thus not eligible for the $100000 prize, so it is not a factor at all. The poaching usually seems to come up whenever we are approaching a milestone and some stragglers are holding it up. Then someone (or ones) take it upon themselves to 'clean up' the trailing exponents. That's why we have been discussing this as if preventing stragglers will prevent poaching; it undoubtedly would do away with most of it. Yes, you can type any exponent you want into worktodo. The GIMPS server will give you an error message if you check in an exponent which is assigned to someone else, but the client will continue working on an exponent it has already started, no matter what. (If not yet started, it will be deleted from the worktodo.) But once you complete it and check in the result, you will get the credit and the person who has the assignment will start getting messages "exponent already tested". This has happened to me on numerous occasions where I picked up a recycled exponent and the person who had it before me (and let it lapse) then finished it before I did. If my machine has already started on it, I have to manually delete it from the worktodo. If it is too far along I just let it go ahead and finish as an early doublecheck. But now I'm drifting off topic (not to mention ranting), sorry. Even though you didn't ask about your 266MHz PII, I will tell you that I have a 233MHz Pentium Pro doing trial factoring. Takes it 19 days to complete an assignment in the current range, but I consider it far from useless. I also have a couple of 300MHz PIIs doing doublechecks, taking about 40 days for 8.9M exponents. A friend of my nephew's is even running DCs on a 120MHz Pentium (his firewall box), takes about 4 months. So your 266MHz machine still has plenty of use... but I wouldn't recommend 1st time LL tests :-)