Printable View
Are there plans for the foresee-able future to migrater the sieve // PRP to the BOINC platform?
Over @ http://boinc.rieselsieve.com we are getting very good Sieve results (almost 10K factors found in a month!).
For Sob sieving I am getting speeds aound 300/301 kps with a Pentium 4!
IMHO, I think it is easy to setup a similiar system for Sob. Only the dat has to be replaced ?
My $0.02
Grtz
C.
We found already 12.000 factors in 21 days.:rock:
This was discussed way (two years) before, and the intention was not to migrate. This discussion was for PRP though.
On the other hand, I agree that there is a nice resource (of users) over there, which we might (should?) make use of... either this way, or the other.
From a quick calculation, Riesel sieve is sieving about 1T per day, which is far from being bad (consider the .dat size!). On the other hand, the forum got more traffic in one month in the BOINC-bug-section bugs than in a year for the whole project.
Are there other ways to get sieve rates of this dimension? SoB has so many users with that much machines who only want to run an out of the box thingy and just are too lazy to sieve. So, I think a small sieve wrapper would do the affair; the ultimate solution would be Version3, of course. No idea about the progress there.
I vote for letting enjoy Riesel sieve the factors they get for the moment; in a month or two, we will see how it works out; and maybe there will be a wrapper one day that helps sieving as well as Boinc does, or even better.
Anyways, I felt like Rieselsieve always had a different policy regarding communication and accessibility of the results, and this BOINC thing seems to go in the same direction.
H.
We are using JJSieve as client program for boinc and it is very solid!
@ Nuri
Did not know this.
I didn't want to resurrect an old thread.
@ HHH:
[quote]
On the other hand, the forum got more traffic in one month in the BOINC-bug-section bugs than in a year for the whole project
[/qoute]
This is true but we got a stable, BETA worthy version for the moment. (version 5.26).
Boinc & the riesel client are very stable (at the moment).
Sorry I do not understand this.Quote:
and maybe there will be a wrapper one day that helps sieving as well as Boinc does, or even better.
Jump over to Stats.rieselsieve.com ... you will see the results.Quote:
Anyways, I felt like Rieselsieve always had a different policy regarding communication and accessibility of the results, and this BOINC thing seems to go in the same direction.
BTW, the stats & boinc thing is almost a one man operation, done by Bryan.
May be its worth rethinking of that option a bit. It has been two years and people might have changed their views, who knows.Quote:
Originally Posted by cedricvonck
Having said that, I feel like we're way way way ahead at sieving (relatively to PRP of course) and if we were going to get a significant resource like this, it would make sense to do PRP instead of sieving.
Nuri I think that Lee is working towards a BOINC version of LLR'ing.
It is true there were lots of posts in the bugs list section of the forum but that is forgivable for a pre-Alpha project. It is much more stable now and I recommend that the SOB team consider the virtues of the BOINC platform. BOINC Riesel Sieve is now so easy to run it is "set and forget".
my 2c
Nuri, I disagree with your opinion. On the contrary, we need more sieving rather than less. We still find more factors per time than we can perform PRP-tests; and we still have a P-1 portion of the project that could be useless, given a better sieve point. And there is joint sieving with lots of work.
Nuri:
>Having said that, I feel like we're way way way ahead at sieving (relatively to PRP of course)
I said this 2+ years ago and I'll say it again,
sieving gets 1 factor per day
prp gets 1 factor per week
do the math.
Initially I calculated u guys should sieve 500T to 2P and you are approaching 2P today, so I stand by my calcs.
If you expect to PRP some factors before 50M, that would make sieving less effective. But it's been 1 year+ since a proof, so keep sieving :)
I seem to remember a famous prime-finder(I forget you) recommending that sieving should be removing twice as many factors as PRPing if it's to be thought of as effective.Quote:
Originally Posted by dudlio
>I seem to remember a famous prime-finder(I forget you) recommending that sieving should be removing twice as many factors as PRPing if it's to be thought of as effective.
That's probably true. Sieving (n=50M) is done way out ahead of PRP'ing (n=~10M) so there is a speculative nature to it. You can figure on a 2:1 break-even point rather than 1:1 because of the unknown. PRP's are real proofs and so they hold more weight. For example, if a prime is found at 25M, then all the sieving 25-50M would be ineffective; thus a 2:1 ratio.
However, there are ~1000 users PRP'ing and only 20 users sieving. With 98% of the project on PRPing, it's doubtful we've ever gotten close to 2:1. If 2:1 is the correct ratio, then we should have 300 people sieving and 600 people PRP'ing.
That's just basic math.
hhh: >SoB has so many users with that much machines who only want to run an out of the box thingy and just are too lazy to sieve. So, I think a small sieve wrapper would do the affair
Agreed. PRP has always had a nice client-server wrapper. It downloads a block, tests it, and uploads the results. There's no reason why we couldn't have a sieve client that does the same thing: reserves a small range, sieves it, and uploads the results.
I've already written a program to submit sieves. With a similar HTTP posting program, we could reserve ranges on Matt's site and automate the whole sieving process. All it would take is a bit of human coordination to make sure that Matt's site can handle the traffic, errors, timeouts, etc.
my 2 с
there'll be much more users that use boinc and don't want to bother with downloading files, extract it and reserve some numbers somewhere
they just configure boinc to run sieve task - and voila! ann files needed to sieve downloads automatically and results goes where they should )))
and having sieve in a project list at http://www.boincstats.com/ can attract curious users. (like me in early boinc years when I test every boinc project. there was 5 of them =)))
What Total Active
Users 3,167 1,625
Hosts 6,578 2,891
Teams 330 238
Countries 83 65
Total Credit 15,453,545
Average floating point operations per second 1,332.8 GigaFLOPS / 1.333 TeraFLOPS
rieselsieve at boinc
the porting of Rieselsieve to Boinc was for sure a good idea. The attention is enourmous. On the other hand, it was a lot of work, too, and some money went there, too, for buying new servers and stuff. That's to say it isn't easy.
My personal feeling is that we shouldn't envie rieselsieve's boinc-success, they need it. The project isn't running so badly after all.
BTW, Death, if you want to do something particular and particulary useful for this project, you can help the combined sieve with Prime Sierpinsky Project at
http://www.mersenneforum.org/showthread.php?t=2666
Yours H.
oh well
when I try to reserve a range in a 2666 topic my posts disappears and I don't see my reservation
665700-666000 completed by Brucifer(not comb)
667000-667065 completed by VJS(comb)
I want to sieve my fav range 666000-666666 that I sieved long ago with 20M dat and this range has high factors dencity.
but somebody just delete my posts or something wrong happend so combined sieve goes to hell with all this difficulties.
*goes back to SoB sieve*
Dear Death, please don't be upset about the system that doesn't post your posts(yet). mersenneforum had a problem with a DOS-attack, and I think that was one of their defenses.
Anyhow, your range is reserved and waiting for you to crunch it. Yours H.
Hey Death I don't think it's as bad as you pointed out...
The range is still credited towards yourself as well as the points. What we have done is redo all lower ranges with a 991<n<50M dat.
There are multiple threads and postings on this topic, currently it is also the dat we are using as well.
Alot has changed since you were active, I'd like to think that the majority of it has been for the better.
Stick with the combined dat you will be the better for it.
Your old range is also archieved here...
http://www.free-dc.org/forum/showthread.php?t=2698
oh thank you hhh for explanation
this was very disappointed as my post just disappered three times )))
ok, I'll do combined sieve after finishing my current range (a week or so) or is there a point to switch to combined sieve right now? and where I should submit 1199000 factors if I switch to combined sieve right now?
vjs, thats exactly what I plan to do )) I remember that was high factor dencity so I want to seieve it again with 50M
thank you for your support. you are great!
and back to boinc sieve
is there posiibility to ADD SoB sieving to boinc riesel sieve? I wonder thats can be done just by replacing dat file... or not?
Just a note from the head cheese at Riesel Sieve. The boinc port was a pain in the butt..but well worth it. It took time to learn boinc thru a learning as you go method, it took time away from our real world jobs, it took a level of humility explaining why our stuff was all stuffed up, it took a lot of money....a LOT of money to upgrade our stuff. A good portion of these funds were donated by our users...the rest was from personal sales of blood plasma, kidneys, and other organs that I simply wasn't using anyway.
After all this...when we are finally settling down into a no hassle day of things going fine....we start with the boinc LLR trials. This will bring more learning, more headaches, more humility as I answer 49 gazillion emails, but hopefully no more money....I'm simply out of spare organs. However, if our project is DoS'd by a mad rush for boinc LLR..I'll gladly sell my car to handle the workload...but our current hardware should be able to handle a few thousand users a day without breaking a sweat.
So...I guess the point is that you should wait and let us go thru another period of hell before you jump on board the boinc bandwagon. We are sister projects and everything basically is the same....so once we iron out all the trouble..you could easily jump onboard with limited problems.
As always....I believe in opensource, open information, and sharing.
Lee Stephens
thanks b2riesel
hope some day boinc sieve will work for both projects....
and also can you say what must be done to add sob sieve to riesel sieve?
i wonder that replacing dat file with combined.dat will be enough?
There would be necessary first some changes to the siever in order to handle k*2^n +-1 in the same time.
Then the sieve-ranges would have to be adapted: SoB/PSP sieves currently n up to 50M, while Riesel is sieving only up to 10 or 20 M, I forgot.
And finally Riesel needs to eliminate A LOT OF k's, as at the moment, the .dat file would be just too huge.
We are not going to see the joint in a near future, but not in a far one either, with a little luck.
H.
Actually I see it differently. Rather than combining +1 and -1 sieving in the same BOINC project, I would create another BOINC project for PSP/SB sieving and leave the current Riesel sieving alone. The client would not have to change. The workload generator and validator would have to change though. The dat would be taken care of and distributed for the new project along with its sieving intervals. The problem is that it would take a lot of resources. Both hardware and manpower. Honestly, the only way that I think it could be done is if Lee decides to do it. Bryan would have to go along with it, or Lee and Bryan would have to choose someone to administer the new project. It may actually be less work for them to run it than to support someone else trying to learn all that they have had to learn to make their project work.Quote:
Originally Posted by hhh
Joe makes a lot of good points there. However, there is one little thing that we haven't mentioned yet...there is a bug...no...a shortcoming in the way boinc handles Work Unit numbering...when we start sieving above 1000T we will run into this error...because the file names will be too long. Bryan may or may not have fixed a work around today while he was playing at home with a snow day. However, this point just shows more reasons why there is still much testing to be done on the boinc front...then...yes..I'd be happy to help you guys out any way I can.
You want a few thousand sievers in a month? No problem. However...in return we ask for help in producing a good, stable P-1 client because personally I believe that is where the boinc effort will show promise. Or I could be wrong... but I'd be man enough to admit it.
You guys need to hurry up and find all of your primes..because we still have a truckload of them to find.
afaik some CPU is more succesful in p-1 and some in sieve, so boinс should choose what method to use.
and ofcourse there's possibility to migrate main sob client to boinc platform. what we need to do this - that's another question.
PS help us find all our primes and all sob sievers switch to riesel )))
To do p-1 work using BOINC it could be a way to use ecm.exe which can be used for distributed work for "p-1", "p+1" and "ECM" work. ECMnet shows that it is possible to use it with a server to feed in new work. So it should be possible to use it similar to jjsieve.
But there are at least two possible problems.
First as far as i know there is no way to have progress information from ecm.exe.
Second i don't know if there is any checkpointing possible.
But the other question is if p-1 is really needed for rieselsieve at the moment.
With the progress you did with sieving i would guess that "p-1" is "usefull" again earliest for n somewhere bigger then 7 Mil. Before that point the number of tests done (even twice for double checks) done within a timeframe by llr is bigger then the number of factors found by "p-1" within the same time.
Lars
who has at the moment neither time nor money enough to play around with
BOINC.
Don't raise the dead.:scared:Quote:
Originally Posted by b2riesel
P-1 should be declared dead. At least soon. There is a reason why riesel isn't P-1'ing anymore. It's just not worth it. Thanks to what? Good ol' sievin'. So, with the Boinc sieving effort, it's not going to become worth the time in near future; perhaps in a quite far one. BUT: I know for sure (and you know my postings about that), that it is way better to boost sieving, not to encourage LLR or PRP too much, and sieve first, better too deep than too shallow, so that P-1 doesn't even become an option.
If you start the LLR client and in half a year you are LLRing at 10M, I will agree that a P-1 client is needed; but unless this happens, you should try to avoid it, IMHO.
I have spoken.
Well this is true but only in a way,
P-1 should never be "avoided" but what we should try to do is make it inefficient.
PRP/LLR vs P-1 vs Sieve vs ECM vs P+1
First we will never find primes without LLR/PRP, when prp testing finds a prime it eliminates the k. So the best way to eliminate all factoring is finding a prime.
I know its hard to believe but it is possible to sieve too deep before testing, fortunately we have fast sieve clients. During the early stages of any project generally k's are sieved to maybe a couple G's then tested to eliminate k's. repeat the process. Currently we are at points were we simply have to sieve as deep as possible.
SoB is also close to the point of where sieve is not worth it, if we only have 1k for example it might be better to only P-1.
there's no my username at http://www.psp-project.de/sieveimport.php
How can I submit my factors?
Oh, well, I just mail them )))
Right, that will do.
psp add ldausch dott de
Lars had problems with his hand, I guess that's why he didn't import you yet.
H.
we need a new dat )))
LTD has already released the PSP/SB combined dat. It is available here.Quote:
Originally Posted by Death
so please update sieve coordination )))