I might have delivered some bad results - Page 3

cheesehead · 2011-11-15, 17:48

Quote:

Originally Posted by LaurV

"Proof of concept:"

I have just reported:
< snip >

Yes, yes, it is easy to falsify mfaktc reports.

Quote:

I should suggest to STOP IMMEDIATELY accepting mfaktc reports until this problem is solved.

Are you familiar with the phrase, "throwing out the baby with the bath water"?

You would have GIMPS stop accepting numerous correct reports merely on the chance that someone submits false reports???

Please consider lowering your alarm level. Relax. Take a deep breath, then let it out slowly. Calm yourself. :-)

Then, if you're still concerned, please describe the incorrectable real-world, or at least actual mathematical, consequences that you fear.

KyleAskine · 2011-11-15, 18:12

Quote:

Originally Posted by LaurV

this is gibberish. Prime95 is also open source (I mean all the sources are available) but it used a DLL to generate the assignment keys which is NOT public. No one knows (beside of George and few of his acolytes) how to generate the assignment keys.

And this is not an overreaction. Maybe most of the bad guys out there did not realize yet (as I did not realized before using mfaktc), but now is public. This is REALLY a major issue.

Come on, it would be trivial to decompile prime95 and forge any checksum you wanted. Please read George's comments on that here: http://mersenneforum.org/showpost.ph...1&postcount=16

One your second point, I fail to see what the bad guys can do with this info. They absolutely could write scripts that generate mfaktc results at random I guess and periodically upload them, but what would they gain from this? It isn't like the LL test where there are actually cash prizes for getting a specific result. I would think that people have better things to.

chalsall · 2011-11-15, 18:16

Quote:

Originally Posted by LaurV

this is gibberish. Prime95 is also open source (I mean all the sources are available) but it used a DLL to generate the assignment keys which is NOT public. No one knows (beside of George and few of his acolytes) how to generate the assignment keys.

You just contradicted yourself.

Quote:

Originally Posted by LaurV

And this is not an overreaction. Maybe most of the bad guys out there did not realize yet (as I did not realized before using mfaktc), but now is public. This is REALLY a major issue.

This "issue" has been known for years.

At the end of the day, as cheesehead points out, the GIMPS primary agenda is to find Mersenne Primes, not factors to Mersenne Prime candidates.

However, most of the serious "firepower" doing TF work are honest, and those (very few) cheaters can be found out by way of statistical analysis of their claimed work done vs. the number of factors they actually find.

LaurV · 2011-11-15, 18:33

Quote:

Originally Posted by chalsall

You just contradicted yourself.

How?

cheesehead · 2011-11-15, 18:39

Quote:

Originally Posted by LaurV

How?

As you said, the DLL to generate the assignment keys is not open source. So, not "all" the source is available to the public. QED :-)

OTOH, the DLL to generate the assignment keys is part of PrimeNet (i.e., the server), not part of prime95 (the client).

OTOOH, the module that generates the checksum on prime95 report lines, which _is_ part of prime95, is not included in the released prime95 source code, so not all of even the prime95 source is available to the public.

(OTOOOH, it's been admitted by George that it's not extremely difficult to reverse-engineer the checksum algorithm.)

Folks who compile prime95 from the open source files get everything except the checksum-generating module, so their reports are flagged as having invalid checksums (but their reported results are still accepted by PrimeNet if they satisfy all other requirements).

Dubslow · 2011-11-15, 19:16

(I know you didn't accuse anybody of anything, but in this case there were 20-25 factors found among the 1250 exponents tested.)

KyleAskine · 2011-11-15, 19:32

Quote:

Originally Posted by Dubslow

(I know you didn't accuse anybody of anything, but in this case there were 20-25 factors found among the 1250 exponents tested.)

That sounds about right, since you would expect around 1/70. Checking new exponents would be a better use of your time than double checking those, IMO

iconized · 2011-11-15, 19:46

@Dubslow: I will help as well, machine runs fine with 2 x 4 GB at the moment. Will also from now on periodically test my memory. Wouldn't hurt if DC applications have some build in memory test. There are projects out there that have no replication/redundancy at all. I guess with all my found factors being confirmed by Primenet, there is a good chance that none of the work was erroneous. The behaviour of my PC became problematic in November besides the unzipping errors, many crashes of Firefox and the display driver.
So if you can PM me 200 lines I will redo them.

@LaurV: All work was done with a GTX 460 Ti, I could almost max it out with one CPU core (i7 2600k) no hyper threading, so I under clocked the GPU a bit.

Dubslow · 2011-11-15, 21:23

That's exactly what I have, without the Ti. One core comes oh-so-close to maxing it out. PM on the way.

iconized · 2011-11-15, 21:36

Eh, I mean 560 Ti. Okay thanks for PM will take a week or so to redo them I guess, will PM you the results. I also upped the couple of tasks below 71 bits to 71.

Dubslow · 2011-11-15, 23:44

Just report to PrimeNet, it's not like I keep track of the results.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Results	ET_	Operazione Doppi Mersennes	627	2021-01-31 18:06
Where have all the TF results gone?...	lycorn	PrimeNet	22	2017-10-02 02:40
PGS Results	danaj	Prime Gap Searches	0	2017-08-14 18:35
I might have delivered some bad results	iconized	Prime Sierpinski Project	5	2011-11-14 11:44
0x results...	Mike	PrimeNet	11	2004-05-23 12:55

2011-11-15, 19:16	#28
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3·29·83 Posts	(I know you didn't accuse anybody of anything, but in this case there were 20-25 factors found among the 1250 exponents tested.)

2011-11-15, 19:46	#30
iconized Apr 2010 Netherlands 21₈ Posts	@Dubslow: I will help as well, machine runs fine with 2 x 4 GB at the moment. Will also from now on periodically test my memory. Wouldn't hurt if DC applications have some build in memory test. There are projects out there that have no replication/redundancy at all. I guess with all my found factors being confirmed by Primenet, there is a good chance that none of the work was erroneous. The behaviour of my PC became problematic in November besides the unzipping errors, many crashes of Firefox and the display driver. So if you can PM me 200 lines I will redo them. @LaurV: All work was done with a GTX 460 Ti, I could almost max it out with one CPU core (i7 2600k) no hyper threading, so I under clocked the GPU a bit.

2011-11-15, 21:23	#31
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3×29×83 Posts	That's exactly what I have, without the Ti. One core comes oh-so-close to maxing it out. PM on the way.

2011-11-15, 21:36	#32
iconized Apr 2010 Netherlands 10001₂ Posts	Eh, I mean 560 Ti. Okay thanks for PM will take a week or so to redo them I guess, will PM you the results. I also upped the couple of tasks below 71 bits to 71.

2011-11-15, 23:44	#33
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3×29×83 Posts	Just report to PrimeNet, it's not like I keep track of the results.