Work exchange?

NBtarheel_33 · 2016-05-23, 09:32

I have ~50 GHz-days in partially-completed double-checks as follows:

23.5% of a 52.3M exponent.
17% of another 52.3M exponent.
5.9% of a 55.8M exponent.
1.5% of a 62.2M exponent.

I do not have the resources to complete all of these assignments, but I would like to avoid wasting the work that has already been done on these candidates.

I wonder if someone would be willing to make a trade: I will send over these partially-completed tests in exchange for 50 GHz-days of crunching on M42429973 (this would be about 34 million iterations, or ~80% of a double-check on this number).

Any takers? (Perhaps if folks are interested in this sort of "exchange", we could keep this thread going with other potential swaps.)

Thanks!

GP2 · 2016-05-30, 07:28

The problem with picking up someone else's work is that you never know if their computer is reliable or not. The Prime95 program works a computer a lot harder for a much longer time than most software, and is very sensitive to the slightest hardware errors. Even one error in a single bit will turn the entire result bad, often without any warning or other indication.

From past experience, a few percent of the results returned (the "residues" of non-prime Mersenne numbers) turn out to be incorrect when double-checked. This varies widely, however: some individual machines turn in hundreds of error-free results, while a few give many or even mostly bad results. This might be the result of overclocking, bad or cheap RAM, letting the CPU overheat, or some other reason.

If there happened to be an error in your 1.5% complete 62.2M exponent, then the person taking over that work would do 98.5% of the work in vain. If their computer has an excellent track record they would be much better off just doing 100% of the work from scratch. Very little extra effort and they would feel much more confidence in the result.

You can go to http://www.mersenne.org/workload/ and just release any exponents that are no longer viable for you. But if you want to continue, you have plenty of time. The main wavefront of double-checking won't reach the 50M range for another couple of years at least. You could just let the testing continue slowly.

2016-05-23, 09:32	#1
NBtarheel_33 "Nathan" Jul 2008 Maryland, USA 5·223 Posts	Work exchange? I have ~50 GHz-days in partially-completed double-checks as follows: 23.5% of a 52.3M exponent. 17% of another 52.3M exponent. 5.9% of a 55.8M exponent. 1.5% of a 62.2M exponent. I do not have the resources to complete all of these assignments, but I would like to avoid wasting the work that has already been done on these candidates. I wonder if someone would be willing to make a trade: I will send over these partially-completed tests in exchange for 50 GHz-days of crunching on M42429973 (this would be about 34 million iterations, or ~80% of a double-check on this number). Any takers? (Perhaps if folks are interested in this sort of "exchange", we could keep this thread going with other potential swaps.) Thanks!

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Official 'exchange of inanities' thread [Was: mm127 is prime, cuz I say so]	kalikidoom	Puzzles	58	2015-10-17 16:18
Best work for the GPU	manfred4	GPU Computing	6	2014-03-18 19:59
How to calculate work/effort for PRP work?	James Heinrich	PrimeNet	0	2011-06-28 19:29
No Work	Pilgrim	Information & Answers	1	2008-01-31 18:53
An exchange with DJones	DJones	Riesel Prime Search	15	2007-08-04 04:43

2016-05-30, 07:28	#2
GP2 Sep 2003 5×11×47 Posts	The problem with picking up someone else's work is that you never know if their computer is reliable or not. The Prime95 program works a computer a lot harder for a much longer time than most software, and is very sensitive to the slightest hardware errors. Even one error in a single bit will turn the entire result bad, often without any warning or other indication. From past experience, a few percent of the results returned (the "residues" of non-prime Mersenne numbers) turn out to be incorrect when double-checked. This varies widely, however: some individual machines turn in hundreds of error-free results, while a few give many or even mostly bad results. This might be the result of overclocking, bad or cheap RAM, letting the CPU overheat, or some other reason. If there happened to be an error in your 1.5% complete 62.2M exponent, then the person taking over that work would do 98.5% of the work in vain. If their computer has an excellent track record they would be much better off just doing 100% of the work from scratch. Very little extra effort and they would feel much more confidence in the result. You can go to http://www.mersenne.org/workload/ and just release any exponents that are no longer viable for you. But if you want to continue, you have plenty of time. The main wavefront of double-checking won't reach the 50M range for another couple of years at least. You could just let the testing continue slowly.