hi
just browsing around the stats of seventeen or bust I fount the user supersecret! just started today and already done more than 1500 tests! What's this, a triple check, or what ;)
just curious :)
Printable View
hi
just browsing around the stats of seventeen or bust I fount the user supersecret! just started today and already done more than 1500 tests! What's this, a triple check, or what ;)
just curious :)
Looks something like that - Louie doing some experimenting I guess.
If you go join the party, current n ~ 41K, and tests take about 2 seconds:cool:
(Louie, sorry if that's messed up anthing you were doing:blush: )
... and whilst on the subject of strange tests, I see that someone is currently testing k=10223, n=13,466,957.
:elephant: :looney:
Ah, I saw that too. It seems like Louie is also betting on getting the largest prime as well. As you know, the largest prime to date has 4053946 digits, and k=10223, n=13,466,957 is slightly larger.
I like that strategy. Who knows, may be we might be lucky enough to hit the jackpot.
Good luck Louie. :thumbs:
wow, that would be the largest prime ever! As most of you probably know the largest was found by GIMPS (am I allowed to mention this, there our competetors? ;)), it is 2^13,466,917-1. Is this the goal, to try and beat GIMPS? I hope we succeed :) Is my guess right, that the smallest possible prime was chosen, that was larger than the current largest? In case anyone is interessted:
current largest prime number (GIMPS): 2^13,466,957-1 (NOV 2001): 4,053,946 digits
this new number 10223*2^13,466,957+1: 4,053,962 digits
(I think, the last figure could be wrong, all the same: only very little larger than the current largest)
current largest SOB prime 54767*2^1337287+1 (2002) 402,569 digits
Hm...
How long does it take to compute one of those tests? Should be counted in weeks, right?
Just a tiny correction.Quote:
Originally posted by eatmadustch
current largest prime number (GIMPS): 2^13,466,957-1 (NOV 2001): 4,053,946 digits
The largest prime to date is 2^13,466,917-1.
My guess is that Louie started with the smallest n available that results in the largest prime.
i created a few new accounts yesterday.
supersecret - is a double check for the secret. it will double check all the numbers under 400k and then it will behave exactly as secret does now. in theory, after it runs it would enable the DC and combined sieve files to be culled up to 400k instead of 300k. may eventually work up to 500k too but doubtfully any higher
garbage - this account is not meant to be run by anyone except one computer run by Alien88. it's a garbage collector that will always give the lowest unfinished test holding back the bounds on the stats page. this is mainly just to avoid the pathologically bad doomsday senerio where a test either keeps getting reassigned and dropped over and over or where someone only returns one block every week for the entire test. in this way, it bounds the total time that any test can take by having a single, reliable system always cleaning up the most neglected tests. for those of you watching the bounds of the stats, this may keep them humming along quicker.
31337 - assigns tests that are larger than the largest prime. meant more for testing than anything else. if someone has a really fast machine and some patience, you can try it out to give others an idea of how long it may take to run tests of this size.
i don't care if people run supersecret. when it hits 400k in a few weeks, it will act just like secret. don't switch too many resources to it of course since the odds of finding primes w/ supersecret (and secret for that matter) are almost 0. it's only an exercise in completeness, a test of the SB client's reliablity, and a way to possibly eliminate the least productive ranges of the sieve.
-Louie
hmm, very interessting, many thanks for the reply :)Quote:
Originally posted by jjjjL
31337 - assigns tests that are larger than the largest prime. meant more for testing than anything else. if someone has a really fast machine and some patience, you can try it out to give others an idea of how long it may take to run tests of this size.
As I would actually like to be the finder of the worlds largest prime (I doubt I would find a prime like this, but who knows? ;)): how long would such a test take (PIV, 2.53GHz), how many blocks would it have?
Also, just to satisfy my curiosity: why the weird name 31337?
Thanks
Ben
EDIT: Is one even allowed to take part? ;)
Maybe you should keep it out of the stats - it makes the extended stats table more or less unreadable...Quote:
31337 - assigns tests that are larger than the largest prime. meant more for testing than anything else. if someone has a really fast machine and some patience, you can try it out to give others an idea of how long it may take to run tests of this size.
I think it's referred to as "leet speech". A time ago when there where BBS' (Bulletin Broadcast Systems) instead of forums, some ppl exchanged letters by cyphers to circumvent the spam/smut filter.Quote:
why the weird name 31337?
31337 (or short: 1337) stands for eleet or rather "elite" and is the most famous "word" of it - and widely used by the "younger generation". ;)
My guess is it should take roughly a week. But don't count on that very much. That might be off by up to 50% (i.e. up to 10 days).Quote:
how long would such a test take (PIV, 2.53GHz)
As a benchmark, an n=16.7m test took 17.5 days on my PIV-1700, when I was crunching GIMPS once.
That's a creative way to deal with the problems mentioned in the "abandoned tests" thread...Quote:
Originally posted by jjjjL
a garbage collector that will always give the lowest unfinished test holding back the bounds on the stats page. this is mainly just to avoid the pathologically bad doomsday senerio where a test either keeps getting reassigned and dropped over and over or where someone only returns one block every week for the entire test.
I'm curious about 31337: It looks as if a n=13.466M would have about 9770 blocks, but the FFTs would have to be longer than the wall of china and I'm assuming each block would take approximately forever (plus/minus a couple days). Is there some known heuristic that says how computing time increases with FFT length? I'm almost curious to try it out on a box, but I'm unlikely to actually finish it -- expect a bunch of dropped tests on that account.
[Edit: come to think of it, this is exactly the scenario where it would be quite useful to have the ability to tell the client "pack up what you've done so far and send it back and let someone else finish it"...]
What if someone actually finds a prime on that account -- do you guys log IP addresses or such? Or how would you know who actually found the thing?
How big is the cache file of a 31337 test?
And does it have a special expiration time?
alrighty, I am running an AMD 1900+ here is what I am seeing.
k: 21181
n: 13466972
cEMs/sec: ~680000
# of blocks: 9773
average block time: 5:42
est. time to complete: 3342366 secs or 38.7 days
cache file: 1.6MB
FYI:
GIMPS says that my 1900+ would take about 12.6 days to complete the same exponent.
Anyway, I am going back to the standard tests.
--
OberonBob
Team ORU
Why not let supersecret doublecheck all tests, and secret only the tests which were not done by SoB? This way you can really confirm a number is composite (two matching residues) and the numbers can be removed from sieving.Quote:
supersecret - is a double check for the secret. it will double check all the numbers under 400k and then it will behave exactly as secret does now. in theory, after it runs it would enable the DC and combined sieve files to be culled up to 400k instead of 300k. may eventually work up to 500k too but doubtfully any higher
A Lucas Lehmer test is more efficient then a prp test the SoB client does. If i understood things correctely, a Lucas Lehmer test can use half the FFT size compared to the test the SoB client is doing. This means that a LL test is more twice as fast. Probably closer to 3 times faster. (Which seems to agree with OberonBob's post).Quote:
As a benchmark, an n=16.7m test took 17.5 days on my PIV-1700, when I was crunching GIMPS once.
If you wan't to find a record size prime you're better of with Mersenne or Fermat numbers.
Here's what I got for the 31337-account with a Pentium IV 2.4 GHz:
k: 10223
n: 13466981
~ 1.01 McEMs/sec
9773 blocks
time left: ~630 hours or 26.25 days
One thing I noticed is that the cache file is named z3466981 and not z13466981 like one would expect.
It's very unlikely but that could lead to some very wrong results if there was an old cache file lying around with the same name...
I noticed that too.Quote:
Originally posted by rosebud
One thing I noticed is that the cache file is named z3466981 and not z13466981 like one would expect.
It's very unlikely but that could lead to some very wrong results if there was an old cache file lying around with the same name... [/B]
Are you going to leave that machine running 31337 ?
rosebud - cache naming scheme is by design. the client can tell the difference between the tests w/o the filename. if you tried to use it for the wrong test, it would delete the cache file and start over.
smh - supersecret is doublechecking all numbers under 400k, including the ones SB originally did. it would take a long time to check higher than 400k and i think doing a check of all the secret tests will be more likely to find a prime than the double checking of our own results for the low #s.
Lagardo - in the unlikely event that someone finds a prime with the 31337 account, i'm confident i could figure out the user by IP address and computer profile. Even if they have a dynamic IP, I could probably find the pool they were in and use their processor description to find them. BTW, before any privacy folks get worried, we only track IP, SB version #, OS version # (or aname data in *nix), and proc ver tag (NOT proccesor ID #). So when I see a test, i can basically guess what class of proc you were using, but in some cases, I only know "P4" or "old Athlon" but I won't even know the freq.
-Louie
ok, on my PII 350 here is what I am seeing:
~ 152000 cEMs/sec
155.8 days :)
I don't really get this. The secret account tests all exponents that were not tested by the SoB client before.Quote:
smh - supersecret is doublechecking all numbers under 400k, including the ones SB originally did. it would take a long time to check higher than 400k and i think doing a check of all the secret tests will be more likely to find a prime than the double checking of our own results for the low #s.
The super secret account will do the same, in effect this means that all numbers not tested by SoB (before secret) will be checked 3 times. Twice by SoB and once by previous searchers.
Whats the problem with letting 'secret' testing all numbers not tested before (this will already takle a long time to complete) and let the 'supersecret' just do a double check on all numbers? Sure this is gonna take a long time, but atleast it's a start of a double check (a few people are putting a lot of effort into sieving the lower ranges!). This way everybody can decide for themself what to test.
Have you followed the discussion about failure rates we had recentely?
No I'm not. My Computer doesn't run 24/7 so it would probably take more than 2 months to complete this test... And since I can't be sure that this k/n won't be removed by sieving anyway, I don't think it's worth it.Quote:
Originally posted by OberonBob
Are you going to leave that machine running 31337 ?
the supersecret IS checking all numbers, not just the ones done by others.Quote:
Originally posted by smh
I don't really get this. The secret account tests all exponents that were not tested by the SoB client before.
The super secret account will do the same, in effect this means that all numbers not tested by SoB (before secret) will be checked 3 times. Twice by SoB and once by previous searchers.
Whats the problem with letting 'secret' testing all numbers not tested before (this will already takle a long time to complete) and let the 'supersecret' just do a double check on all numbers? Sure this is gonna take a long time, but atleast it's a start of a double check (a few people are putting a lot of effort into sieving the lower ranges!). This way everybody can decide for themself what to test.
Have you followed the discussion about failure rates we had recentely?
several of the k's were started below 400k, by SB and those are being done by supersecret just like the secret tests. the 400k limit is just a temporary limitation based on how i set it up. it can be raised later, but it won't be an issue for several weeks.
i realize that this will represent a triple check of some of the very low numbers, but in most cases, those numbers even if they have residues will take very little time to redo. if you still have contact w/ any of the previous searchers, they are welcome to email me residues for comparison and that would help me avoid triple checking. the whole idea is just to get a pair of matching residues so that we can eliminate the number for good... and eliminate it for sieving too.
and yes, i did follow the failure rate discussion. GIMPS failure rates are about the same as ours. i have seen several numbers being reported late that were already reassigned and finished that give inconsistence residues. however, they are the minority... i have also assigned quite a few test primes and none has ever been mistakenly reported composite. there was also a sever error (or something dave did w/o telling me) a few months ago that assigned a number 5 times. all tests came back; all were the same residue. in fact, i just ran a script and so far 1660 tests from supersecret have been reported and we have 1660 matching residues. :) error rates tend to increase with test size, but these are still encouraging numbers.
-Louie
It would be interesting to see your statistics on error rates, Louie. Wblipp pointed out that this information could be incorporated into a resources allocation model. My guess is that the probability of an error should be proportional to the duration of the test, so we should see error rates per test go up with time, as is also the case with GIMPS. At least with some data, we would have a basis for predicting the probability of finding a prime doing double checks as opposed to checking new numbers.
I've been doing supersecret for a couple of days now, but it has stalled several times. I'm getting these errors (v1.10):
[Wed Jun 11 22:12:41 2003] connecting to server
[Wed Jun 11 22:12:45 2003] logging into server
[Wed Jun 11 22:12:56 2003] couldn't report to server [can't log in], retry in 60 secs [error: 0]
[Wed Jun 11 22:13:56 2003] connecting to server
[Wed Jun 11 22:14:02 2003] logging into server
[Wed Jun 11 22:14:03 2003] couldn't report to server [report denied], retry in 60 secs [error: -3]
[Wed Jun 11 22:15:03 2003] connecting to server
[Wed Jun 11 22:15:13 2003] couldn't report to server [can't connect], retry in 60 secs [error: 0]
Any ideas on what's wrong? One test takes ~6minutes, am I too quick? :D
Definitely not, one of my PCs does a supersecret test in 3-4 minutes... :smoking:
But I don't have a solution to your problem. If it is v1.10, then I can tell that it ran well for over a day by now. But according to the stats, the PC stopped crunching one hour ago. :cry:
I'll have the answer to that pitty discovery on Friday when I get access to that machine again.
How many supersecret tests are gonna be done? Wouldn't a double check mean that each test that is performed would only have about 1% (This has been indicated as the chance of a prime being reported non-prime) of the chance that it had of finding a prime the first time around? I assume this means that just going on would almost always be the best allocation of resources until double checking can be done in 1% of the time that the current tests can be done. This makes sense to do until the tests take 50.4 minutes (still a good way off) at least that is the amount of time they would take at this point to obtain this ratio on my computer that takes ~3.5 days for a current test. Then procede with double checking as the project continues but always kee pthe resources focused in the place where they have the best chance of finding a prime. After all its not the end of the world if we miss a prime as long as we find another one for that K its all good. als ohave any of the residues from the supersecret tests not come back matching yet?
Does anyone know anything about this?
Looks like I was the only one to continue with a 31337 big test (under my own ID).
28433*2^13466977+1 completed earlier today :D
Knowing my luck it's already been factored out by sieving (I haven't checked).:rotfl:
At least it is shown in the stats. :cheers:
I just checked: no factors. ;)Quote:
Originally posted by MikeH
Knowing my luck it's already been factored out by sieving (I haven't checked).:rotfl:
btw. how long did it take (time and blocks)? Was there a big KcEM/sec increase?
I'll get back to that PC on Sunday, so I'll let you know then. Elapsed time won't be too representative anyway, because I've done the odd bit of sieving on that PC.Quote:
btw. how long did it take (time and blocks)? Was there a big KcEM/sec increase?
Just a short note from the supersecret front:
We've passed n=300k today! :|party|: