100M exponent idea

MrOzzy · 2008-10-16, 12:52

We all know with the current technology and software it will take sevral years to test an 100M candidate.
Since it takes so long for someone to complete the test chances someone gives up someware down the line are rather high.
People also will use mutiple cores and waste a lot of cpu time to the overhead trying to finish the LL test as fast as they can. It is a lot more efficient testing multiple candidates at once. So why doesnt GIMPS split up the work? For example someone runs a LL test on an exponent for 1 month. The results of this test are checked in and uploaded to the GIMPS servers. Now the person checking the exponent can continue testing it or pick something else.
If eventually a prime is found everyone working on a specific exponent will be credited according to the amount of work they contributed.
I'm not sure this is technically possible, but if one single computer can continue where it left off another computer also can.
It also will allow double checking someone else work before a complete exponent is finished.

Mini-Geek · 2008-10-16, 12:59

Technically this is possible, but would require transferring files that are several megabytes, so probably won't be done.

jrk · 2008-10-16, 15:27

Yeah the smallest 100M digit Mersennes would take about 40 MB of storage and bandwidth to store an interim file.

henryzz · 2008-10-16, 19:03

40mb would take me about 20min to upload i think currently although in the future i could see a plan like this working

potonono · 2008-10-16, 23:18

Is that theoretical 40 Mb compress, or uncompressed?

axn · 2008-10-16, 23:25

Quote:

Originally Posted by potonono

Is that theoretical 40 Mb compress, or uncompressed?

Theoretically, the intermediate values of the LL test should be fairly random and thus compress poorly (less than 5% reduction).

/SWAG

retina · 2008-10-16, 23:43

I cant see how this "idea" would encourage people to run 100M numbers in any way. Someone runs, say, the first million iterations then uploads the residue file to somewhere, then another person downloads it and runs another million, etc. How does that help towards running it any faster? How does that stop people being bored with it? There are still no final results for many years and people just see numbers (and large data files) coming and going for years with seemingly no conclusion. What happens when a final test is returned with a bad residue? After all the hard work everyone does and one bad CPU/MOBO has ruined the result for all! Who provides the huge bandwidth and disc space to store the interim files? That is not going to be cheap.

joblack · 2008-10-16, 23:46

There would be also the problem whom to give the 75.000 USD (from the EFF price).

Does prime95 support the nehalem 12 (/24) Core architecture. It should be possible to shorten the time for one 100m digital number under 1 year?!

jrk · 2008-10-17, 00:00

Quote:

Originally Posted by potonono

Is that theoretical 40 Mb compress, or uncompressed?

The big numbers in the interim files are not compressible at all (well, it could theoretically be compressed very small, if your compression algorithm includes a Lucas-Lehmer function to decompress it

but that would be defeating the purpose of storing the interim file in the first place...).

Just for fun I compressed some interim files with gzip and bzip2.

Gzip inflates the files by 0.016%.

Bzip2 inflates the files by 0.44% and takes 6 times as long as gzip to do so.

ATH · 2008-10-17, 00:11

Savefile for my 46M exponent (13.87M digits) takes 5.76Mb, so that is around 41.5% of the digits in bytes, so if this holds a 100M digit exponent would take 41.5Mb.

Batalov · 2008-10-17, 00:47

Dudes, the save file simply contains a 64-byte header with the status data and then the (rotated) residue modulo your number (that is p bits). That means that the size of the savefile is 64+p/8 bytes (=40551Kb for the lowest number), and it is surely incompressible (because it is essentially random noise).

2008-10-16, 12:52	#1
MrOzzy Apr 2008 Antwerp, Belgium 111001₂ Posts	100M exponent idea We all know with the current technology and software it will take sevral years to test an 100M candidate. Since it takes so long for someone to complete the test chances someone gives up someware down the line are rather high. People also will use mutiple cores and waste a lot of cpu time to the overhead trying to finish the LL test as fast as they can. It is a lot more efficient testing multiple candidates at once. So why doesnt GIMPS split up the work? For example someone runs a LL test on an exponent for 1 month. The results of this test are checked in and uploaded to the GIMPS servers. Now the person checking the exponent can continue testing it or pick something else. If eventually a prime is found everyone working on a specific exponent will be credited according to the amount of work they contributed. I'm not sure this is technically possible, but if one single computer can continue where it left off another computer also can. It also will allow double checking someone else work before a complete exponent is finished.

2008-10-16, 15:27	#3
jrk May 2008 3×5×73 Posts	Yeah the smallest 100M digit Mersennes would take about 40 MB of storage and bandwidth to store an interim file. Last fiddled with by jrk on 2008-10-16 at 15:27

2008-10-17, 00:11	#10
ATH Einyen Dec 2003 Denmark 3⁵·13 Posts	Savefile for my 46M exponent (13.87M digits) takes 5.76Mb, so that is around 41.5% of the digits in bytes, so if this holds a 100M digit exponent would take 41.5Mb. Last fiddled with by ATH on 2008-10-17 at 00:11

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
100M exponent FFT size?	NookieN	Information & Answers	3	2017-09-04 22:11
Windows 10 in Ubuntu, good idea, bad idea, or...?	jasong	jasong	8	2017-04-07 00:23
how to submit manually a job with exponent 100M which is done by mfaktc?	fairsky	Information & Answers	17	2013-09-16 19:49
overclocking an i7-2600 to finish an 100M exponent in less than a year :)	emily	Hardware	4	2013-02-28 20:11
What minimum exponent would give 100M digit prime?	odin	Software	7	2010-04-18 13:57

2008-10-16, 12:59	#2
Mini-Geek Account Deleted "Tim Sorbera" Aug 2006 San Antonio, TX USA 17×251 Posts	Technically this is possible, but would require transferring files that are several megabytes, so probably won't be done.

2008-10-16, 19:03	#4
henryzz Just call me Henry "David" Sep 2007 Cambridge (GMT/BST) 2³·3·5·7² Posts	40mb would take me about 20min to upload i think currently although in the future i could see a plan like this working

2008-10-16, 23:18	#5
potonono Jun 2005 USA, IL 193₁₀ Posts	Is that theoretical 40 Mb compress, or uncompressed?

2008-10-16, 23:43	#7
retina Undefined "The unspeakable one" Jun 2006 My evil lair 2²×1,549 Posts	I cant see how this "idea" would encourage people to run 100M numbers in any way. Someone runs, say, the first million iterations then uploads the residue file to somewhere, then another person downloads it and runs another million, etc. How does that help towards running it any faster? How does that stop people being bored with it? There are still no final results for many years and people just see numbers (and large data files) coming and going for years with seemingly no conclusion. What happens when a final test is returned with a bad residue? After all the hard work everyone does and one bad CPU/MOBO has ruined the result for all! Who provides the huge bandwidth and disc space to store the interim files? That is not going to be cheap.

2008-10-16, 23:46	#8
joblack Oct 2008 n00bville 1011011000₂ Posts	There would be also the problem whom to give the 75.000 USD (from the EFF price). Does prime95 support the nehalem 12 (/24) Core architecture. It should be possible to shorten the time for one 100m digital number under 1 year?!

2008-10-17, 00:47	#11
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 3⁶×13 Posts	Dudes, the save file simply contains a 64-byte header with the status data and then the (rotated) residue modulo your number (that is p bits). That means that the size of the savefile is 64+p/8 bytes (=40551Kb for the lowest number), and it is surely incompressible (because it is essentially random noise).