Mersenne: PrimeNet Stats Updated
PrimeNet's updated CPU stats chart is at http://entropia.com/ips/stats.html The last chart update was mid February. There are a handful of days that reached 800 gigaflop rates, and a nice view of the rate recovery from the v17 bug. A linear trend is no longer the best fit. Regards, scott Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
RE: Mersenne: A few questions
>How large will the exponent be for a 10,000,000 digit prime number? To be a 10,000,000 digit prime number the exponent must be at least 33,219,281 (which also happens to be a Mersenne candidate). >Has the prime number that was found a week ago been announced on this >list? >I.E. What number was it? It hasn't been announced yet... but from what little information that is available, i.e. The Oregonian newspaper article, the exponent must be =at least= 6,643,859. Eric Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
RE: Mersenne: A few questions
CORRECTION TO LAST MESSAGE! The exponent 33,219,281 happens to be a Mersenne prime candidate! Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Re: Mersenne: LL & Factoring DE Crediting
Brian J. Beesley writes: >thereafter. I think George's idea was to encourage people to do the >appropriate amount of trial factoring - if you didn't do enough, and >proceed to run the LL test, you lose the credit because you should This doesn't fit with my understanding of how Prime95 works. I though that the program automatically chose the factoring depth using a formula that weighs the chance of finding a factor against the time that would be required for the LL test. If there is a way for the user to control the amount of factoring, then it is news to me. David A. Miller Rumors of my existence have been greatly exaggerated. Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Re: Mersenne: Status estimate
At 19:03 06/28/1999 -0400, you wrote: >The Status estimate of the chance that the number you are testing seems to >be off by a factor of e, based on Wagstaff's estimate. > >+--+ >| Jud "program first and think later" McCranie | >+--+ Yeah, I was wondering about that thing...if we don't even know if there are finite or infinite Mersenne primes, how does it generate that number? Mine says my odds are about 1 in 60,000 or soI'm doing an LL test on a number in the 719 range, and back when I was testing numbers in the 400 range the odds were much better...around 1 in 45,000. --Peter Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Re: Mersenne: LL & Factoring DE Crediting
Hi all, At 09:59 PM 6/28/99 +0100, Gordon Spence wrote: >if a factor is later found for a Mersenne number or the >Lucas-Lehmer result is found to be incorrect, then you will "lose credit" >for the time spent running the test." > >It is on always struck me as odd I must admit. >At the time the test was done it obviously WAS required, so why the >subsequent penalty when we make advances and can factor much deeper? The explanation is really simple. It saves me from having to keep track of too much data (the bad results and the ones that a factor was later found). However, since the v17 bug, I've had to modify my procedures anyway and give credit for those incorrect results. I suppose it wouldn't be too hard to do the same for factors found and LL results proven incorrect. Regards, George Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Re: Mersenne: LL & Factoring DE Crediting
At 09:59 PM 6/28/99 +0100, Gordon Spence wrote: >The GIMPS home page explains the following > >"Finally, if a factor is later found for a Mersenne number or the >Lucas-Lehmer result is found to be incorrect, then you will "lose credit" >for the time spent running the test." > >It is on always struck me as odd I must admit. It isn't very clear, but the purpose may have been to discourage bogus positive results, because they will eventually be double-checked. +--+ | Jud "program first and think later" McCranie | +--+ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne: Status estimate
The Status estimate of the chance that the number you are testing seems to be off by a factor of e, based on Wagstaff's estimate. +--+ | Jud "program first and think later" McCranie | +--+ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Re: Mersenne: LL & Factoring DE Crediting
On 28 Jun 99, at 21:59, Gordon Spence wrote: > "Finally, if a factor is later found for a Mersenne number or the > Lucas-Lehmer result is found to be incorrect, then you will "lose credit" > for the time spent running the test." I obviously can't answer for George. PrimeNet works a different way, Scott credits any work submitted in good faith, whatever happens thereafter. I think George's idea was to encourage people to do the appropriate amount of trial factoring - if you didn't do enough, and proceed to run the LL test, you lose the credit because you should have done more factoring to start off with. Otherwise you maximize your credit by simply running LL tests & ignoring factoring altogether. The problem with this approach is that goalposts shift. Different processor types have different LL test:factoring performance ratios, in particular the "plain Pentium" (Intel Pentium "classic" and Socket 7 MMX processors) are rather poor at factoring compared with most other processor types, and processors with high core:bus speed ratios are going to be (relatively) better at factoring because of the lower memory bus load. Consequently the appropriate factoring depth varies to some extent depending on the mix of processors in use. > > It is on always struck me as odd I must admit. > At the time the test was done it obviously WAS required, so why the > subsequent penalty when we make advances and can factor much deeper? > I agree. However, to take the point to a ridiculous extreme, finding a factor saves running a LL test - so why can't I have credit for finding 54,522 (small) factors in the range 33.2 million to 36 million, thus saving (very approximately) 54,522 * 8 P90 CPU years LL testing? The job ran in an hour on a PII-350! Somewhere in here there is a compromise. I'd suggest: (a) you should lose _double_ credit for a LL test if the result is proved incorrect, or if a factor is found in a range which you claim to have checked; (b) a table should be published of the suggested factoring depths for various exponent ranges. If you submit a LL test result, then someone finds a factor less than the table depth at the date you submitted the result, you lose the credit; otherwise, you keep the credit; (c) however deep you factor, you only get credit for factoring to the table depth in the current table. BTW I don't really understand why PrimeNet doesn't credit you for results obtained when the manual testing pages are used instead of the automatic system. I suppose Scott is trying to promote the automatic system, and I'm in agreement with that sentiment, but it still seems a bit strange to me. Regards Brian Beesley Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Re: Mersenne: A few questions
On Sun, Jun 27, 1999 at 09:08:15PM -0400, George Woltman wrote: >Assuming they respond that the claim >is all in order, then we should be able to announce shortly thereafter. I'm keeping my fingers, toes and hairs crossed :-) Just too bad nobody else has participated in my guess-contest... That means I will be the sole winner! Hooray! Seriously, though, perhaps Slashdot will post the news when the exponent is official. I've already submitted it three times... /* Steinar */ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne: RE: Mersenne Digest V1 #584
[On the off chance that no one else has replied to this since it was received...] On Sat, 19 Jun 1999 12:55:46, Jeff Woods wrote: Starting at about M13, you see that there are indeed islands: Given the strong linearity of "log(exponents of mersenne primes)", it is not surprising that the averages of consecutive pairs or triples will also be linear. Indeed, taking the "centers" of the islands as data points will actually produce a stronger correlation than the original data. If I go back and average consecutive exponents (well, the log of the exponents, anyway), the correlation with a straight line improves from .9925 to .9935 (compared to the original data). If I get to throw away the ones that don't "fit" according to some arbitrary criterion, it may improve marginally. [Note: using your choices actually reduces the correlation to .9849!] So perhaps we should remember Mark Twain: "There are three kinds of lies: lies, damn lies, and statistics." -Shaun Quantum Mechanics: The dreams stuff is made of Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne: LL & Factoring DE Crediting
The GIMPS home page explains the following "Finally, if a factor is later found for a Mersenne number or the Lucas-Lehmer result is found to be incorrect, then you will "lose credit" for the time spent running the test." It is on always struck me as odd I must admit. At the time the test was done it obviously WAS required, so why the subsequent penalty when we make advances and can factor much deeper? regards Gordon Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Re: Mersenne: A few questions
At 07:08 AM 6/28/99 +0100, you wrote: > > I *suspect* that in light of GIMPS' success, he is likely looking much > > higher than we are now (and has been for some time, again as a > guess). He > > knows our current program cannot top P > 20,000,000, so I suspect he's > > above that range, perhaps by a good margin. It may be that he breaks > GIMPS > > record(s) again someday. > >Are you sure about this? By no stretch of anyone's imagination. As I said, i don't know David from Adam, and have never talked to him, even by Email. My speculation was a SWAG, and could be 180 degrees from the truth. Your own information about his ability to get CPU time does tend to contradict my guesses, which were based on the heresay I noted below. >George told me that he'd asked David S. to >verify the newly-discovered prime, but that David was having >difficulty getting access to sufficient CPU time & would only be able >to run in the background, taking approx. 4 weeks. Several others of >us would be able to do an independent double-check run on a 6 million >range exponent in that time, e.g. I could use MacLucasUNIX on my 533 >MHz Alpha 21164 & complete in 4 weeks (ish). Which is why I contacted >George, to offer my services. > > > I do know that the last time George expanded > > Prime95 to hunt up to 20MM (up from 3MM), that George sent several > residues > > in the upper ranges to David for confirmation, and that David was able to > > confirm SOME of them rather quickly (faster than a Cray could do the > > calculations on the spot, so they were already tested). This to me > > indicates that David is searching the numbers far above our top potential > > range, especially since a Cray can test such numbers in about a week or > > two, as a guess. > >Well - for testing the algorithm you don't neccessarily have to run >to completion - I've also run quite a number of exponents between 10 >and 20 million to 400 iterations for the purpose of cross-checking >results. And I'll be doing a lot more when I get sufficient memory. While not CERTAIN, I'm fairly sure that the test values David confirmed were of "last iteration" residues. >Verification of 3021377 took DS 8 hours on "his" Cray. A "square law" >seems reasonable (run time = #iterations * time per iteration, time >per iteration is O(n log n) where n is the FFT size) so, 20 million >being approx. sqrt(50) times 3 million, run times of 50 * 8 hours = >2.5 weeks would be expected - assuming unlimited access to the whole >system. For exponents >33.22 million (10 million digit numbers) he'd >be well into months per test. YOWCH! How long on a P-III/450, or an Athlon (K7), then? Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne: M617 Factored
M617 Factored - M617 has been completely factored by the Special Number Field Sieve (SNFS). It was previously known that M617 = 59233 * 68954123297 * c171 The p5 was found by Riesel in 1957 and the p11 found by Brent and Suyama in 1981. The c171 is a 171 digit composite factor given by c171 = 133162933696720252644109076239739315294641129598\ 571214674268232878869403201703608966454713865163\ 367575359986404237817149731676559992850220509804\ 718874844608767326097407871 On June 26,1999 it was found that c171 = prp51 * prp120, where prp51 = 1577519781152253854956475324210064781272294056\ 44601 prp120 = 8441284558692203904329647194935146184026400044\ 2557390949549814809580375481579281829664306776\ 0548829906291314807571121271 The factorization of M617 was 'Most' wanted by the Cunningham project[1] and is the smallest number of the form 2^n-1 whose complete factorization was not known. That distinction now goes to M619. The sieving was done by a group of 19 volunteers starting May 7, 1999 and finishing on June 16, 1999. A total of 15986534 relations was collected requiring about 1.2 Gbytes of memory in uncompressed ASCII format. The resulting matrix was 1563691 x 1566436. The linear algebra and square-root phases were done at Centrum voor Wiskunde en Informatica (CWI) by Peter Montgomery. Acknowledgments are due to the volunteer sievers Pierre Abbat Ricardo Aguilera Brian Briggs Gary Clayton David CrandellConrad Curry Kelly HallPhilip Heede Jim HowellSkip Key Alex Kruppa Samuli Larvala Don Leclair Ernst Mayer Thomas Noekleby Henrik Oluf Olsen Marcio de Moraes Palmeira Guillermo Ballester Valor Paulo Vargas Special thanks to Bob Silverman, Peter Montgomery, Alex Kruppa, Don Leclair and Ernst Mayer. Also to CWI, the Department of Computer Sciences at the Technical University Munich and the School of Mathematical Sciences at the University of Southern Mississippi for the use of their computers. M619 is currently sieving by SNFS. It is on the Cunningham projects 'More' wanted list. If you would like to donate some of your CPU time visit [2] and download the siever for DOS or linux. Source code is available on request to compile on other platforms. You will need at least 21 Mbytes of free physical memory. [1] http://www.cs.purdue.edu/homes/ssw/cun/index.html [2] ftp://ftp.netdoor.com/users/acurry/nfs Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne: Re: 10,000,000 digit prime
The 10,000,000 digit prime would have an exponent of over 3010299.956, or 3010300 which is found by taking (log 2 * 10,000,000) J. Zylstra Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Re: Mersenne: Linux VFAT question
On Sun, 27 Jun 1999 11:45:37 -0400, Pierre Abbat <[EMAIL PROTECTED]> wrote: > > But the fact is that the performance of my mounted hdd win98 (vfat) > > partition is very low. All my normal linux (ext2) partitions work with > > normal performance. > > cd to various directories, then type vdir|wc -l in each. This tells you > how many files the directory has. The more files, the longer it takes to > sort them. vdir -U lists the directory without sorting it, which is a > little faster. I too find that performance on fat/vfat partitions is low compared to ext2 partitions, but I don't think that's too surprising: the kernel is understandably optimised for ext2, and FAT partitions are hardly the most efficient of filesystems in the first place. To be honest, I don't see a lot of difference in performance on fat partitions whether or not mprime is running. > mprime just writes once every 30 minutes, so whether it's running in the > Windows drive or the Linux drive is inconsequential. As to the swapping, > the swap daemon runs at 12 naughty, so I don't see why mprime at 20 nice > would bother it at all. On one of my machines, disk performance appears significantly worse with mprime running than when it's switched off. On my P200MMX, performance on a Quantum Fireball ST4.3A drops from about 8.25MB/s to 4.9MB/s (measured using hdparm). However the performance of the Seagate drive on the same machine and the Fujitsu drive on my PII-266 is barely affected by mprime. Still, I'd rather lose a little on disk performance than waste over 96% of my workstation's CPU cycles :-) One write every 30 minutes is hardly going to trouble things much. I'd far rather have the machine doing regular disk writes than risk losing several weeks' (your uptime may vary) work if mprime is not terminated neatly. -- Robin Stevens <[EMAIL PROTECTED]> Merton College, Oxford OX1 4JD, UK http://www-astro.physics.ox.ac.uk/~rejs/ (+44) (0)1865: 726796 (home) 273337 (work) 273390 (fax)Pager: 0839 629682 Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm